But those days are over as AI chatbots have improved massively over the last couple of years. From a user point of view AI chatbots now feel a lot more like chatting to a real person with relevant and useful answers. From an implementation point of view, it requires an entirely different approach.

Three of my biggest learnings from the implementation are:

1. Accept that you have less control over the precise answers given by the chatbot.

2. Your data is much more important than you think and you might need to structure it entirely different from what you think.

3. Don’t ignore AI adoption as an integral part on the implementation plan.

The implementation process took much longer than anticipated, but it will also take only a fraction of the time next time around, now I understand the setup and what to look out for.

Hubspot have rolled out a large number of AI features over the last two years, but all are in beta mode and many appear poorly tested in production with real teams, which causes some headaches for us trying to use them.

Their customer facing chatbot is no exception.

Because the chatbot is very untested at scale, Hubspot have continued to change the setup over the last year - for the better to be fair. But that meant that each time I tested to see how well it worked in the last year, everything had moved around, were called something different or seemed to work differently from the last time. During the recent implementation they continued these changes and improvements. So you should know that before you get started, to avoid some frustrations.

The implementation was a steep learning curve, and at times it felt like I was doing it all blind folded. But the end result was impressive, so in this article I will try to give you all my best advise, so that you can implement it with a lot less hassle. I’ll go over the technical setup, how to work with your data in various ways and how you ensure that adaption and change management are not forgotten. Along the way and at the end you’ll find lists of elements to keep an eye out for that I hope will be helpful.

Enjoy the article!

Prepare to make the technical setup without any guidelines

Hubspot does not yet have a catalogue of good cases of the chatbot working effectively over time, which means that it can be difficult to get useful recommendations from others’ experiences. This applies to both the technical setup but also to choosing the right data strategy to fit your specific data and context.

Over the course of the implementation I therefore had to try out multiple methods based on how other chatbots - both AI powered and the more traditional chatbot - are commonly set up. That led me to dead ends over and over again. Thank good for a very patient client who was prepared for pivots when needed.

Even something as simple as googling guides for the setup was not fruitful as all guides I did find, were outdated with yet another Hubspot update - that was even the case for guides that were only a few weeks old.

However, in the end I did manage to get the chatbot working, and we were finally on to testing.

I am reluctant to give precise guidelines to the setup here, as I fear they will be outdated in the near future, so I will opt to give you a list of elements to consider when you get to set up one up for yourself, but be ready for changes to have happened meanwhile.

1. Find the best way for you to add your data. For me it was using website url’s as the client already had a very rich website, with deep content that covered most of the knowledge we needed the chatbot to know of. But for you it might be using only Knowledge Articles or to use Knowledge - which is not the same as “Knowledge Articles” nor the same as “Knowledge Base”. Yes, Hubspot should really work on their labels in general.

2. Short answers are not a good way to organize data. This was initially one of the ways I planned to provide data to our chatbot, but firstly, short answers are not given very high authority over url’s, and short answers get really difficult to manage, as they can’t be sorted into categories or folders, but are just displayed as a long list of questions. Never use short answers if you can find other ways to add the data in a more rich format.

3. Take notice of whether the implementation script works for both your chatbot and your Hubspot Forms - for us it didn’t work, and we had to write a custom script that enabled both to work across multiple websites.

4. Set up chatflows - these are useful if you want to differentiate what happens in the chat for different pages or sites or at different times during the day. Even if you just have one uniform way, you still need to set up one chatflow and make sure if is connected to the Live chat and that it uses AI

5. If you have a large website with 50+ pages, make sure to organize your url’s into categories, so that they are manageable over time. If you add an url and activate related urls, you’ll end up with a big pile that can be difficult to manage later. However it will also take time to upload one url by one, so make time for that. Take notice that the url search function doesn’t work, so it can be difficult to find url’s if they are nested as related urls.

6. Make sure to add your Hubspot script to the right level on your website. We had some browsers reject it and mark it as a marketing cookie, which required a cookie accept up front for users located in EU. This might also be a problem caused by your cookie pop-up provider.

7. Set up human handover to your help desk and make sure to test if tickets actually land in help desk, as only setting up the handover in the Customer Agent section is not enough on it’s own.

Your data sources and quality are essential to the quality of answers

Perhaps quite naively, we started the projects by defining a very constrained and modest goal: To have the chatbot answers the 20 most asked questions that customers support got in their support email. That turned out to be a completely wrong approach and a hard lesson that caused many hours of wasted work.

We extracted over 9000 emails and analyzed them into categories and themes, discussed them with client back and forth, prioritized them and then we got started inside Hubspot.

Failure about to happen

Hubspot offers many ways of adding content for the chatbot to use, but it can be difficult to get any gist of what works better, and how data is prioritized, so I simply had to get to work and test out loads of ways until i found a way that worked. I started with two data sources: a scrape of their website url’s across 3 markets, which gave me 2000+ url’s and then knowledge articles with our selected 20 answers. And now came the exciting test to see if we could get correct answers to the minimum requirements of 20 questions. But the chatbot failed massively. Hardly any of the answers were correct and the chatbot continued showing me that the sources used to generate answers came from url’s and not the 20 carefully crafted knowledge articles. In the end it became clear that knowledge articles had absolutely minimal authority over the url’s. Especially short knowledge articles. Long ones might be assessed differently relative to urls, but that will require additional testing.

Pivots in process and goals

It resulted in a major pivot and the abandonment of our goal of the 20 precious questions. We realized that using the url’s to launch the chatbot and have it - theoretically - answer any questions was a better way forward. But it also required the client to accept that we could not upfront promise which questions it would do well on and which it wouldn’t. Hence making the scope between client and agency entirely intention-based, rather than metric-based.

So new goal, new plan and new approach. And a reminder to approach AI projects with an open mind, fewer hard KPI’s and with a willingness to change direction, plan, deadline and accept that that is all a part of the process, when working with this technology that are still relatively new and that evolves faster than we can say pivot!

Less control over answers

Through years of working with marketing and communication I’ve encountered lots of clients who are refining their brands and want to perfect everything from messaging, icons, digital assets, and tone of voice. However when you implement an AI powered chatbot that uses generative AI to compose answers, you are required to let go of some of that control.

Control is different from old style chatbots to Hubspots AI chatbot

With older type chatbots with predefined questions and answers, that was never a problem. Same goes for the copy on the website, in emails, to a certain point how your customer support employees converse, but for the Hubspot Chatbot that control is much more difficult.

You should not be too discouraged about that though. EU law prescribes that AI cannot interact with users without them knowing they are conversing or interacting with AI. In addition some studies have found that users actually prefer that AI chatbots have a neutral tone of voice in contrast to a very branded tone of voice. The neutrality can emphasize to users that they are chatting with an AI chatbot and not a human.

Take notice of EU law

I should also mention that we went with a solution for the introduction message from the chatbot that said “Hello, I am an AI chatbot and still under training, so you might experience that I make mistakes from time to time. How can I help you?” which also made it clear to users what they were dealing with.

But be prepared to let go of some control of wording and let the Hubspot chatbot answer in it’s neutral build-in tone, which we found to be polite, neutral and close to what most human customer service agents use in most organizations.

Chatbot adoption for your customer service team will determine if the implementation is worth the investment

Working with AI Adaption daily I knew I needed to get the customer service team onboard with the AI chatbot project. I have had plenty of experience in bringing in both AI and other new technology changes into different organizations and first hand seen what happens when the right and affected teams are not included. Safe to say it will be uphill or downright a complete loss of your investment.

The purpose of the chatbot

First you need to identify what the purpose of - in this case - the chatbot is. Is it to terminate the customer service team, to move their resources to more complex queries or VIP clients, to improve response time and response quality, or something entirely different?

Be honest with yourself and with the agency doing the implementation. The purpose will greatly shape how you approach change management and testing during the project phase.

Unfortunately I do often still hear management teams say that the purpose is to cut salaries and employees to a minimum. To be clear that was not the case for this particular projects. And I must say, that I believe it is the worst reason to implement AI in most cases. The purpose should be innovation, better products or better customer service, development of new revenue streams or a number of other great reasons that build and expand your business.

IKEA case example

A really good case is IKEA who, with the introduction of an AI chatbot, had 8500 employees retrained into product development. The staff handling customers support issues had developed deep knowledge about the customers, the products and where gaps were present and IKEA gained an increased revenue of $1.4 billion with this brilliant shift of focus of their dedicated staff 1. This should serve as a prime example to all of us.

What many employees fears, are that with any AI solution they will be let go, and so if the organization does not address this, and are clear about what their new role will be, or how their resources will be reallocated to different tasks or priorities, who can really blame them for resisting the change! I want to also acknowledge that I am aware that the resistance towards AI vary greatly based on where you are in the world, the nature of the industry, and communications inside the individual organization. So adoption should be adjusted to your specific context.

My five cents on AI adoption

These are my five cent on how to ease the adoption when you introduce an AI chatbot into an organization where customer service are mainly done via email, forms and phone up until now. Naturally it will need adjustment based on your concrete case, but let this inspire you:

• Involve the affected team as early as possible, let them ask all the questions they have. Don’t rush it. Show real empathy, and skip the I know better attitude.

• Explain what the purpose of the chatbot is and how it will affect their jobs, role and priorities.

• Underline what AI chatbots are good at, and where they lack skills. Don’t paint a picture of the technology as being perfect or entirely able to do precise what a human can do. If you can’t clearly explain the difference yourself, you have some reading up to do.

• Let them volunteer to be ambassadors and help out with the process and quality testing. Don’t force anyone who are still not convinced. You will need them both to contribute with knowledge and you need them to understand the technology’s weak spots and how to refine it ongoing.

• Ask them questions along the way about the processes, customers and products. It will enable you to make a better set up for just them.

• Make sure to train them thoroughly along the way, and let them get their hand dirty early so they are soon familiar and can start refining the chatbot before the end of the project phase - which will also increase the likelihood of finding technical or process problems early on.

• Encourage their ideas and critique, as well as their ability to solve problems themselves. This will give them a deeper understanding and much better background knowledge to advocate the AI chatbot to their close colleagues.

• Define roles and responsibility together with the team and plan for what happens when the chatbot answers incorrectly, what the procedures are, how it can be avoided. Resist the urge to assign blame to a human for something done by AI. The chatbot will make mistakes, so plan for it and work on minimizing it going forward instead.

You might also like this article

Results after one month and future expectations

We started out with one goal of having the chatbot answer correctly to 20 carefully selected questions. Along the way we pivoted and agreed that the goal should rather be to get it live, clearly be able to identify knowledge gaps and have processes in place to correct those. As we saw the chatbot working on conversations, we realized that in the first month it was able to answer two in three questions to a satisfying level. The rest were almost entirely due to missing data available to the chatbot. That made us set another goal for the year of maintaining the answering rate at two in three questions, even as the client was going into their high season where the number of inquiries increase massively.

With the great results we saw in the first month we also agreed, that a goal of reaching nine in ten questions answered would be ambitious, yet realistic to aim for within two-three years, given that they keep adding new data for the chatbot to utilize.

Impressive results

In general I was quite impressed at how well it could answers even complex questions as long as data was available somewhere on the website or uploaded inside Hubspot. So even if the price per conversation is a little steep, the infrastructure and speed at which it can be implemented, as well as the quality of conversations I would recommend it to anyone using Hubspot already today. I will also recommend that you negotiate the price of AI credits which are used for conversations, with Hubspot. They are usually able to give you a good price when you commit to more AI credits monthly.

Final advise before you start the implementation

• Prepare your client and/or yourself for pivots, changes to the setup, test and failures and avoid thinking this will just work after you push “publish”.

• Your data can come from many places, but depending on the specific combination of data sources you have available, Hubspot will assign value to them differently.

• Prepare for a long test period and for the chatbot to be a constant object of refinement. This is not a project you implement in a month and then leave alone for the rest of the year. Set up a plan for continuous monitoring, improvement and quality operations.

• A consultant cannot effectively quality check your answers, as they lack the deep knowledge of your brand, products and procedures. Make sure to involve the customer support team as part of the implementation process.

• In general I highly recommend involving the customer support team to ensure adaption is possible and effective. You need to make sure they are on board, understand the purpose and how their resources will be used once the chatbot starts taking over some tasks. Without proper adaption the project will be doomed from day one.

• Before you purchase any AI credits, you can test it for free with the included AI credits. But be aware that once you run out, it will stop answering questions, but remain live. Hubspot will not give you any notification of it, so keep an eye out for it manually by opening conversation one by one, or simply make sure to buy credits, which also means you’ll get notifications when you are about to run out.

• Prepare for less control. Often we want to control our responses as much as possible, but just as with people, we can’t provide a manuscript and expect answers to fit precisely into a box - we need people to think and come to conclusions that are aligned with our business, and for an AI chatbot we need to provide the right amount of data to support correct answers, while we monitor to secure quality and provide more knowledge to improve outputs.

Lastly I want to emphasize that this article and my review in no way was supported nor encouraged by Hubspot. This was my honest experience in implementing the AI Chatbot, both with its great results, but also with the unnecessary hard way of getting there. Let’s hope they take notice and make it easier for us soon.

Let’s work together

Hi! I’m Christina and I help organizations across sales, marketing, and operations understand where AI can drive efficiency, improve decision-making, and unlock new innovative opportunities.

I focus on turning AI into something useful, through clear use cases, strategic prioritization, and solutions that can actually be implemented.

My approach is practical and business-focused. No hype, no unnecessary complexity, just a clear path from idea to impact.

Catch me on LinkedIn

Thank you for reading!

There are more ways you can show your support:

❤️ Tapping the heart below helps my articles get discovered

👀 Follow my account to see my content in your feed

💬 Comment what you think of this post

🌟 Become a subscriber if you are not already

Subscribe now

In my daily work as an AI consultant I work with all employee and management levels of companies actively working on transforming their companies and engaging is building more AI competences. Through that work I see what really happens out there, from small to large corporations. What struggles they encounter, what new needs are occurring and how they with little or much success move forward in the new world order where AI runs the show and most just do their best to keep up.

A year ago training to increase AI adoption was mostly centered around compliance and general technology understanding. Sometimes training included workshops in specific tools and other times more general AI news were on the agenda. Alle of these activities had one thing in common - They were all designed for a broad audience and always below middel manement level.

But that is drastically changing!

When I work with companies now it is clear that the most skilled are often lower in the hierarchy and the least skilled are the once who define strategies, set the direction and who are meant to lead with great example.

However, to build strategy and direction in a meaningful and efficient way, you need to know what you are talking about. If not, the stratagies are fluff, aspirations, and without any root in reality.

And the hard truth is - people who know what they are talking about and who have engaged and build AI competences - they can see right through it!

Fortunately many of the leaders I meet have realised that they must up-skill in order to lead the conversations, and so I am increasingly training c-level and mid-level managers - even more often that employees at lower levels.

Training C-level

And the training of the top of the hierarchy are much different to what we commonly know as AI training. First of all, in my experience almost everyone at c-level will only participate in 1:1 training. There is a lot of pride to take into considerations with this type of training, as the participant is in a vulnerable position where they need to admit what they don’t know in order to learn. And while that requires a certain type of person and the building of a good relationship, the relationship cannot be too good nor close. If there is no relationship they won’t open up and share what they know and don’t know, they won’t trust you with the work they do or let you in close enough for you to give them useful advise and training. And on the other hand, if too close, they can feel like they are letting a work relationship know of their weaknesses, or even worse that those weaknesses might get spilled to some of their peers or competitors.

You might also like his article

Training Mid-level

Training of mid-level managers are also different. They usually work in small groups for some topics and individually for other topics. Pride can be a problem in some organisation dependent on the culture, and not at all in other organisations. The aim of the training is for them to be able to define tactics, measure result and to communicate value of AI and obtain budget and resources upstream, while communicating individual value, deflate doubts, build confidence, clarify policies and allocate time, budget and energy, and encourage the team to learn.

For mid-level managers it will also often makes sense to build community where they can share experiences, tips, guides and challenges they face, while c-level are rarely to never interested in joining any communities - even if they are solely made up of same-level individuals.

AI Adoption is one of my absolute favourite activities, regardless of level they each have their own challenges to be solved or worked with.

For anyone interested in AI Adoption and how to approach it whether in small, medium or large organisations I offer a few tips that I find useful in my own day to day.

1. Most AI training need to be specific to the function or department

2. Most AI training cannot be given with real effect to more than 20 people at a time

3. AI training is not just compliance and guardrails

4. AI training should be a continuous activity, not a check mark on your to-do list

5. Consider if AI training should be mandatory

6. AI adoption requires patience, human sensibility, and requires people who are likeable

7. AI adoption is a major culture shift to any company

8. AI adoption should include both compliance, security, workshops, individual training, work on mindshift, Q&A, leadership backing, budget and time allocated

9. In most cases AI adoption must be concrete and almost never abstract

10. AI adoption and training is the single most important activity you can do to advance overall AI capabilities today. Much more important than tech or data.

Thank you for reading!

There are more ways you can show your support:

❤️ Tapping the heart below helps my content get discovered

👀 Follow my account to see my content in your feed

💬 Comment what you think of this post

🌟 Become a subscriber if you are not already

Subscribe now

Read more of my work

However…In my daily work with AI adoption in primarily B2B organisations, it is clear - There are still a tonnes of companies who are:

• Not actively using AI

• Where employees are unsure of the companies stance on tools like ChatGPT and CoPilot

• Where leadership have a hard time defining a strategy - often because they simply lack knowledge, AI vocabulary and a sense of possibility and limitations themselves.

And as time goes, these organisations cirkel into a downward spiral where every day makes it more pressing to prove AI business cases, but where it simultaneously gets harder and more embarrassing to admit that you have not even begun at all.

But rest assure! You are not alone. The only embarrassing thing is to continue not doing anything. Sooner or later someone will realise it and by then it will become a real problem.

So let’s get started! I here offer you a practical framework to get started. You don’t need any particular skills, nor a team of AI engineers or a defined 5 year AI strategy. All you need is determination to get started and willingness to learn.

A 9 step practical AI adoption plan

I have developed this plan to fit any type of organisation and it has evolved over the years as I have gotten wiser and as the world has changed. I am sure there will be future versions as well, but for now the 9 steps will get you started on solid ground.

The 9 steps are not to be done in any particular order, several of the steps must be in a continuously loop, some will overlap and some are activities require your attention every 6 months. I will make sure to let you know when to do what, in a clear straightforward manner.

I have numbered the steps solely for the purpose of better readability, and to suggest an order you can use. But again, each case and company is it’s own.

Step 1: Assess your situation

If you are a the mere beginning of your AI journey, start by making an initial assessment of your situation. You should look at the four elements: people, data, strategy and experience & projects.

Return to your assessment again in a few months, when you have completed a few more steps and your understanding have deepened. Adjust the assessment accordingly, as it will come in handy when you start to implement AI more organically into your strategy.

People (1/2 a page)

Do you have employees who are digital and for sure or perhaps know something about AI? Are your employees curious or opposed to using AI? How many are using it and how do you think they use it?

Data (1/2 a page)

How good are you at structuring data about customers, products, processes and so on? What systems do you use to organise data? And is data a part of your business model or more of a supporting process for sales fx.?

Experiences & Projects (1/2 a page)

Have anyone already completed or begun any small scale AI related projects? How is it going? Do you have any idea for projects you think could be interesting? What is holding you back?

Strategy (1/2 a page)

Do you already have a clear and concrete AI strategy? Is it clear enough for people to start using? Do you have defined AI policies and guidelines?

For this first assessment you should not comprise a 100 pages report, interview every department head and examine every single proces and system to get an overview. Set a boundary for yourself to make this in 1 hour and about 2 pages. If this assessment step becomes too big, you will never find time to do it and you are non better off. So keep the task limited and realistic.

Step 2: Run pilot projects to gain momentum

You should very early on begin working with a few small pilot projects. Don’t wait for the perfect timing, don’t choose the one project that has the biggest impact..all in all don’t wait.

I often meet business leaders who want to start with an AI project that will completely transform their business - but that is the worst possible idea.

Instead select your first pilot projects based on the following criteria:

• Something you can think of today

• Something that only require one or two people to implement

• Something that takes maximum one day to implement

• Something that solve one small non-business critical task

You don’t want to launch a customer facing chatbot without any experience and risk something as critical as customer support to a failed first project.

The purpose of pilot projects is to better understand what possibilities and limitations AI has, to test the implementation process of an AI project and to see small scale value delivered fast.

I recommend you run pilot projects in every department before involving teams in implementing bigger projects, to promote excitement, confidence and lay the foundation of experience first.

Step 3: Identify your AI ambassadors

When you run pilot projects look out for individual or teams that are particularly excited and skilled. These will be your AI ambassadors that will make AI adoption much easier in day to day operation.

Make sure not to appoint AI ambassadors that don’t want the role upon them. If they don’t have any genuine interest, they will probably not do a good job and even worse they will possible stall or even oppose any AI project implementations.

Make sure to provide extra training for your ambassadors, career advancement opportunities, and support their extra efforts to help you move faster and more smoothly on with adoption across teams.

I will also advise you to not take these people for granted and expect them to use their free time to learn, while you harvest their newfound expertise. I have seen this way too many times, and it always results in them finding other employers who appreciate their skillsets properly. We are looking into a future jobmarked where AI specialist are scarce and companies will soon start fighting for their endows, so make sure to appreciate and reward them in order to keep them inhouse to your continious benefit.

Step 4: Define your AI Policies

When working with clients of both large and smaller organizations I see two mistakes happens often when it comes to AI policies.

Too complex and heavily based on compliance

Some companies rely too much on compliance and perfection when they write up their AI policies. This means that until the very complex policy document have been perfected, no one knows what they can and cannot do. It also often means that the policies are so complex they take too lang to read, even more to understand, and they are seldom revised and hence quickly outdated.

They also might be so complex that many employees are simple too scared to overstep the policies that they simply don’t implement anything at all. If they even read the document at all. You see - lot’s of problems with this approach!

Non exciting Policies

Other companies don’t understand the importance of having guidelines for employees and the majority of companies report that they don’t have any policies or that they don’t know if they do. This is also a problem. Not having any guidelines signals that AI is not a priority, it makes what, how, and when uncertain for employees and that will almost always result in shadow use of tools that can compromise the company.

The golden middle way

In most cases the best solution is something in-between.

Start by making simple life rules or guidelines. They should fit into one page. I suggest 5-8 rules that outline the companys approach and position. It could be something like:

• Don’t paste customer data into chatbots

• Document which AI tools you use and what you use them for

• We are open to AI tools and encourage you to try out productivity enhancing tools

• Consider if an AI tool is biased towards anyone before using it

• and…

Do not attempt to make these guidelines overly complex. It will take you too long to finalize them and your employees will have nothing to guide them until you do.

Next, work a version 2

I suggest you revise your AI policies every 3 months. Yes that is often. But because everything moves fast in AI these years, we need to be on our toes and keep adjusting them to stay relevant and useful for you and your employees.

The next version will naturally be more specific and be more connected to compliance matters. You will perhaps have seen some use that are not aligned with your values or business that you want to include. And this version is the right time for that.

Finally your AI policies should not be a legal contract with the purpose of placing blame if overstepped. Instead make it a useful tool, a living document that can guide your company in day to day AI adoption.

You could also consider having an anonymised channel where employees can ask questions about rules, compliance and policies.

Step 5: Facilitate AI training

As soon as you have decided that AI adoption is a priority, you need to provide AI training to your employees. By now many will already have tested out a few tools themselves, but expect that some will not have tried any.

Provide training in the following:

C-level training

C-level and other management levels often require 1:1 training specifically customized to their specific needs. In reality many leaders are not comfortable about putting their eventual lack of AI knowledge on display, to remain in a position of leadership. Depending on the company the need for 1:1 training can be relevant to several layers of management and absolutely necessary to foster a solid knowledge base to those defining the strategy, tactical plans and AI guidelines down stream.

General AI training

As part of communicating your stance on AI and as a general purpose training include training for everyone regularly. This does not mean once or twice a year, but preferably as often as every month. This will of course depend on the type and size of the company, and if you are just getting started or have been doing some implementation already.

Specialised training

Specialised training covers the training that are given to special functions, teams or individuals and that are special to their daily work. This should be given often for AI ambassadors as their require a deeper set of skills than others, for other the specialised training is less often. Design this to each team as you discover needs or in alignment with their role and how much AI affects their work.

Compliance training

The reason I list this as a separate type of training is, that many companies provide compliance training, and then think they have completed training. That is wrong. Compliance training will not teach your employees how to use the tools, so make sure you have training that teaches how in addition to compliance training.

Step 6: Share knowledge

I addition to the practical training, you must also share knowledge. I’d argue for many this should be the first step, however it will depend entirely on the size and nature of your organisation.

In small organisations I suggest sharing knowledge in person as often as once a week. This will be a very strong signal that you prioritise AI. Anyone can share knowledge, and it could both be AI news, a project or pilot that someone did or even just showing a case from a company similar to your own.

The important thing here is to start talking, inspiring and to create a language and vocabulary that enables people to participate without being singled out or without any or much experience.

For larger organisations in person sharing will look different. Make sure to create space at townhalls, afterwork sessions and even digital channels or boards where people can share their AI projects or knowledge. Encourage sharing experiences and reward those that do. Make sure that knowledge sharing is easily accessible to anyone who wants to participate. You could also very easily create a digital AI library on the intranet dedicated to share knowledge, but it should not stand alone. Make sure you take charge and set a good example as a leader, as you cannot expect people to seeks out knowledge if it is not signalled as important by management.

Leadership and management must make sure they are also sharing how they use AI in their work, and I can guarantee that will be the biggest amplifier of encouragement for employees. I suggest appointing a small team to take lead on the knowledge sharing format, but with open invites for anyone to share.

Step 7: Measure your AI adoption level

If you are serious about leveling up on AI adoption you have probably also considered how to measure if you are getting better.

This is no easy task.

First off, you must be confident in what your goal is before you can measure it. And you must have a strategy on how you will measure it. So far the best way is still a questionnaire. It’s easy to compare, produce - even if people are not fan of questionnaires, they do prove to be an efficient tool. However you should consider other methods as well, but they will depend on what your goal and methods are.

If you don’t need large scale quantitative metrics, consider measuring the effect of individual or teams related to a projects. One way is the questionnaire, but also consider if a small interview or talk could suffice or in slightly larger teams, you could ask people to interacts with an interview chatbot to discuss what and how they benefit from AI solutions.

Make sure to measure broadly and more narrowly at the same time for metrics relevant to each.

Lastly I would recommend you adjust your metrics often - so while the first 6 months you might need to measure how many use AI solutions, later you might need to measure how effectively they use AI, or if they are advancing their processes, creating new possible revenue streams, or using AI to do tasks that they didn’t have time to perform before.

Know that measuring AI adoption is a difficult task. Finding the right balance between too few and too many metrics, how often, what to measure and how to use the findings from it is a big task in itself. Make sure to create time for it.

Step 8: Identify opportunities across organisation and prioritise

When you have completed a few pilot projects it’s time for more serious business. Time for the value creating and more advanced projects.

But how do you identify projects in general, and how do you access which ones add the most value?

First of all, your approach is naturally depending on your company landscape. For some it makes sense to begin in departments with high adaptability, for some it makes sense to start with departments that are production heavy and for some in departments where you can easily see opportunities for AI optimisation.

For each department start by mapping the processes. The most simpel way to do that is to write down a list of overall responsibilities they have. Break each of them down to smaller tasks, so you end up with simple tasks such as “make a pitch deck” or “write a meeting summary”. When you have all tasks written down, access how much time is spend on each of them.

Before selecting which tasks to continue with I advise you select based on one or more of the following criteria:

• Takes a lot time and is somewhat repetitive

• Is boring or trivial

• Can easily be solved by AI, possibly in combination with some automation

Which of the three criteria you assign most weight in your decision is up to you.

You now have a list of tasks that can be assisted or solved by AI, and a very actionable list of projects to include in your AI project roadmap.

Step 9: Implement projects, evaluate, measure and repeat

It is now time to start implementing projects. Some projects might be small and can be completed by one person - often the person who usually owns the task before AI. While other projects require a more complex team of experts, in some cases outside resources, data scientists or engineers and designers.

For the bigger projects make sure to set the right team and don’t pass on too complex projects to a team that does not have the right qualifications. The risk of failure will grow exponentially with the lack of qualifications.

For each projects I always recommend that you access what you will gain from the project beforehand. In some cases you get a happier team with fewer boring tasks and that is your KPI. And in others you save hours spend and reduce time from leads to sales completion. You should also make sure to describe just how you are going to measure the effect of the projects at the beginning, as that can help guide decisions along the way.

Lastly when a project is completed, gather the team and evaluate the projects. Too often we skip this part because it is boring, and people are on to the next thing, but because AI is moving and develop fast, and your company and team as well, it is important to stop for a second and evaluate what your learnt during the project. The leanings will serve you very well in the following projects and have a massive potential to save you valuable budget, time and hurdles going forward.

Get the workbook

I know that getting things done can be difficult in a busy day. To help you work through these steps efficiently and structured I am creating a workbook that takes you through each step, one by one. The workbook will include thorough guidelines, exercises and examples of each step to remove any doubts and encourage more precision from the get go.

Get notified when the workbook is released in January. For anyone signing up before the end of 2025 I will send you the workbook for free. If you are already subscribed, send me a private message or write “9 step Workbook” in the comments to get your copy.

Thank you for reading!

There are more ways you can show your support:

❤️ Tapping the heart below helps my content get discovered

👀 Follow my account to see my content in your feed

💬 Comment what you think of this post

🌟 Become a subscriber if you are not already

Ok so let’s start with clarifying some terms. Vibe-coding is not a protected term and the definitions are many. But most people should be able to agree that vibe-coding is the act of building a digital product that commonly would require coding skills. However, with vibe coding you can now build it using natural language just like you would talk to ChatGPT or any other language model.

So what does that really mean? It means that you can build websites and whatever apps you feel like on your own, without any experience in coding. And isn’t that insane!

I think so!

Ok, let’s go!

Over the last couple of months I have developed 5 apps and website in Lovable, and from that I collected a list of all my learning, so perhaps you can skip those hard lesson and move a bit faster than I did.

I divided the list into five parts:

• Determine your app/website complexity level

• Select a plan & credit use

• Planning vs. freestyle

• Integrations & security

• Tips on how to design picture perfect(-ish)

Determine your App/website Complexity Level

The first thing you need to do before anything else, is to think about and determine how complex your app or website is. This will help you choose the right build strategy, how much you need to plan, how many credits you need - and it can potentially save you many hours spend.

Use this as a guideline:

These things add complexity, so the more you can tick off from this list, the more you need to plan and stock up on credits.

Features that add complexity:

• Customer facing

• Published on App store or Play store

• Integrations with systems where data is shared between the two systems

• User levels and handling of user data & login features

• The design need to be pixel perfect to an existing design guide

• Complex database structure, fx. with user data

Low complexity features:

• An app only you will use

• A simple website with no integrations

• Freedom to design even if only 90% brandguide compliant

• Simple integrations where you show (but don’t exchange) data, fx. an instagram feed

Select a plan & credit use

Free version won’t get you far

With lovable you get five credits for free every day to start building. A credit is not a set amount of work or prompts, but it depends on how much and complex you build.

But those five credits run out fast. If you are an extremely patient type who can build for 20 minutes and then wait to continue the next day for another 20 minutes, you’ll be fine and you can use it completely for free. But most of us are normal people - and in that case you should just go for the 100-200 credits straight away.

How many credits you need

I used anywhere between 30-300 credits per app/website to get to a first version ready to launch. It really depends on your plan and the complexity of what you’r building.

None of the solutions I build took me more than one week to complete, so consider just topping up credits and not committing to a locked plan for 6 months - yet!

Planning vs. freestyle

If you are not too limited on credits or time, I recommend that you try to build something with no plan and another something with a plan.

For me it was interesting to see where Lovable took me when I didn’t give it too many precise details and often the suggestions it made were really good, so don’t underestimate a good “going with the vibes” session.

On the other hand, and especially if you are building something customer facing, complex or if you are already a professional say UX or UI designer or you work as a programmer, you need a plan. Both to optimize time, building, credits used to build and to end up with something that you and your client have agreed to.

Try to still keep some details open, you’ll be surprised at what improvements it can offer for especially the UX. After all it is called vibe-coding, so vibe with it!

Integrations & Security

Security

The second you start adding integrations, or user logins to your app your complexity level is up. I believe there are still some concerns about security, most recently I have read about prompt injections that could be a security issue for Lovable apps, so make sure you need it before you start adding fancy integrations amass.

If you are in a position where you have access to a developer I do recommend consulting them to help you build a structure that is secure. If not, then Lovable do help you with highlighting security issues and it can also help you fix them. But it requires patience and credits.

I am however confident that Lovable will keep on improving this, because in the end it will be their downfall if they don’t. So I remind myself that no-one had heard of chatGPT 3 years ago, so I’ll allow Lovable some more time to get on top of this. Until then I make sure to be extra intentional with what and how I build.

Fixing issues

When you go on to publish your app you likely get a security warning for more complex apps. Don’t worry, Lovable does - as mentioned - help you fix those issues. My advice is to fix them one by one. Often I found that attempting to fix all at once was too big of a task for it to handle. It can be done, but fixing security issues might take about the same time it took to build the rest, so brew a bucket of coffee and put on your patience hat!

Tips on how to design picture perfect

If you are building a website or app for a client you will most often experience that they want it picture perfectly aligned with their brand guidelines. Which is fair of course.

But first off, tell them that one of the benefits of building in Lovable is that it is cheap and fast. And in some cases that means only 90% perfect.

Second, I tested a few ways to get to design that was pretty dang close to the design file and so far my best way is to do the following:

1. Design your app or website in Figma or similar tool

2. Realign with the client that this is what we are aiming for, but we might only get to 90%

3. Tell Lovable what pages you need, and the full website or app structures, but ask for empty pages

4. For each page, tell Lovable how many sections you’ll need besides the menu and footer

5. Cut your design into smaller pieces in Figma fx. → menu, hero, text section, footer and so on.

6. Upload design pieces one by one and ask Lovable to design them like the reference image. Make sure to also upload image files, as Lovable otherwise will generate and use credits to create new images. And make sure to mention what section you are uploading.
   Example: About page, add a section 2, design it precisely as the reference image uploaded. (And then upload your reference file and the image/s it should use.)

   I have tried uploading full pages as well, but it ended up ignoring and adding random things as well, so I stuck to one-two sections at a time.

7. Sometimes you need to tell it twice, so mention again what page and section and ask it again to design like the reference.

You could try to add a reference file that also has px measurements on it and see if that will work. I haven’t tried that myself yet, but it will be on my own list for next time.

There is also an edit feature where you can go in and edit a page, by manually removing a section fx., but each time I tried it, it randomly made other changes when I saved it. So I keep clear of that for now, and stick purely to prompt editing.

Wishlist for Lovable’s next update

With my current experience, I have collected this small wishlist that I hope will be included in the next updates:

• Known security issues like prompt injections obviously need to be fixed

• It would be great if security issues like user database issue don’t occur in the first place

• I could use plug and play integrations build in for the most commonly integrations/connectors OR better guidance on how to do them in collaboration while keeping them secure

• The edit feature needs to either work or be removed

• I’d like to upload a full page design file at once and have it output more precisely on that or even a Figma integration, so I dont need to download and upload files, but instead I can directly reference a page, section or element.

My Verdict → WOW!

Lovable is still young, they were founded in 2023, re-named Lovable in December 2024 and today less than 12 months later, it is enabling us to build without any or much coding experience. And that is thoroughly impressive!

Lovable is not perfect, obviously the security issues is first priority and the building could be made even better. But when I compare this to other AI driven website/app builders out there, this is miles and miles ahead of them in usability, precision and usefulness.

Even for builders who have absolutely no coding experience, this is a tool that will help you build, but also learn you a thing or two along the way about things like API keys, databases and how functions work.

So my final verdict is just as the headline here says WOW! This tool has impressed me and I cannot wait to see where they take it over the next year.

Thank you for reading!

There are more ways you can show your support:

❤️ Tapping the heart below helps my content get discovered

👀 Follow my account to see my content in your feed

💬 Comment what you think of this post

🌟 Become a subscriber if you are not already

And knowledge sharing is the perfect recipe. Kind of like boiling an egg, really not that difficult, as long as you know where the pot is and what an egg is.

So here is my list of 10 reasons why knowledge sharing is the biggest enabler to get you started and continue to foster widespread AI adoption.

1 - Build an AI vocabulary

Sharing knowledge about AI news, tool etc. is the best way to create a common language of AI. A language that starts with simple terms like prompt, agents and training data - and later more complex terms. Enabling your employees to talk about AI is the first step in having them think about AI and how it could be used to their benefit in their work.

2 - Enable leadership to think of AI as part of the company strategy

If management don’t know how to talk about AI, they won’t be able to include it as a serious part of the strategy. But often leadership will be the last to admit they don’t have much knowledge in this area, so a common sharing space is a perfect way to indirectly train them without them having to loose face.

3 - Create a space of AI curiosity

Sharing knowledge in some format will both signal to employees that AI is a priority, but it will also help spark curiosity across teams. Keep the format open and allow for questions and discussions. This is where new ideas often begin.

4 - Discover your AI critics and learn from their perspective

Observe the crowd when you share AI knowledge. Some critics will be verbal about their criticisms and others will only show it with their disinterest and disapproving looks. Make space for criticisms, concern and resistance. And find ways to work with the problems areas that are raised. If possible, then involve the critics to get more insights about their perspectives and use that to find answers, take a stance or to refocus what topics are shared the next time.

5 - Discover your AI ambassadors and benefit from their enthusiasm

Knowledge sharing sessions will also bring out your passionate ambassadors. Make sure to identify, acknowledge and involve them in AI initiatives. They will be your biggest supporters, they will test and highlight faults, they will bring a plethora of new initiates and they will spread knowledge, and experiences up and down the chain ongoing if given the time and space to do so.

6 - Share experience and mitigate risks

One of the most common reasons why projects fail is because we miss to look at the mistakes and experiences of others. Sharing projects - good and bad - will mitigate some of the risk of new projects. But make sure to foster an environment where people feel safe in sharing bad projects for it to work the best.

7 - Clarify leaderships views on AI and control the narrative

According to a 2025 study 7 out of 10 employees use AI tools that are not approved by management beforehand. If leadership and management don’t control the narrative, define rules, and share views it is to be expected that people will take action into their own hands. People will use AI regardless, so management must share concrete perspectives, and use knowledge sharing sessions as a was to communicate point of views.

8 - Harvest knowledge and build open it

None of us can keep up with the speed of AI. And thousands of new tools are launched everyday - and so we must make sure we work together to benefit from knowledge across teams. Even an unsuccessful project in one department can trigger an idea in another. Sharing sessions or forums can be a valuable space to collect knowledge. Not just of what has already been tried and implemented but also about the limits and possibilities of AI technology that can be useful to continuously build upon.

9 - Minimize rouge behaviour and shadow tools

With a forum for sharing, exploring ideas and somewhere to ask questions, you create an environment of creativity, openness and curiosity. And that in return will minimize both shadow tool use and behaviors that can cause potential problems. Which takes me to the 10th and last item on this list..

10 - Exemplify compliance

Sharing examples of how other departments used AI is really effective at showing concrete examples of broken compliance or considerations about how to make sure we comply to rules, law and ethical use. Many companies have seminars where they present theoretical frameworks and imaginary examples of compliance issues. But in my experience it can be hard for people to apply theory back into their own role and into their own tasks with just theory. Real examples are much more effective at that.

And that’s it!

And that’s it! Every organization I’ve seen succeed in achieving high AI adoption has mastered knowledge sharing to a tee - because without people’s understanding, knowledge, initiative, and excitement, AI will never truly unfold and reach its full potential.

Thank you for reading!

There are more ways you can show your support:

❤️ Tapping the heart below helps my content get discovered

📧 Forward this to a friend

🌟 Become a subscriber if you are not already

• “If I just work harder I will get noticed”

• “It’s my mangers job to keep track of and understand what I do and its value to the company”

• “Why can I say things over and over again, but they never listen to my expertise”

Now you might think I sound like a petty person who just wants the attention of others. But this goes deeper than that. Not getting noticed at work was really a catch 22 for me, because I wanted to be able to get a pay raise and advance in my career, but I also disgraced putting myself out there and attracting attention telling about my work.

Me, me, me.

I used to think that colleagues who put themself at the center and who openly told about their accomplishments were suckers, attention seekers, ass kissers and narcissists. But over the last couple of years my perspective has dramatically changed.

If you are still reading, I will tell you what I changed.. yes “I” , because this change will only come from you, so keep on reading and I’ll spill the beans so you can change too. I promise you don’t have to kiss anyone’s ass.

Not a book review, but this kickstarted everything

A few years ago I read a book and it completely changed how I looked at work. That books was “The No Club: Putting a Stop to Women’s Dead-End Work”. I will recommend this book to everyone women and men - but in particular all women, because we are just statistically not as good as men in this particular area.

Very briefly - because this is not a book review - The book is written by four professors who had struggled with overtime, stalled career advancement and being guilting of not saying no to any task at work, which in return gave them even more overtime work. They go on to develop a framework that highlights the value of their tasks at work and how you can start working with that, as well as strategies for saying no to so-called non-promotable tasks.

This book kickstarted something for me. Dare I say, it added new and exploding connections in my brain, and gave me a new perspective on not just work, but so many things.

Okay. Again, read this book. It’s worth it!

7 things that I changed

1 - I realized I could not change others, so I needed to change myself

I used to blame others for not owning their responsibilities. “I told them so many times, why do they keep on making the same mistakes” or “It’s his responsibility, so I need to wait until he does this before I can do that” . Today when I look back at some of the things I considered barriers, I can’t believe that was me. Horrified but also smiling because I found a way out.

I am not even sure what the triggering event was, but one day I just understood that I can’t change others efforts, work, perspective or opinions. If I want things to change I need to change my own perspective. I need to move around that mountain, because it ain’t budging! Change how I see things and how I do things.

And that gave me back the power. It’s impossible to change others, but I can pretty easily change myself.

And so I did.

2 - I mapped my company’s currency

One of the exercises from the book “The No Club: Putting a Stop to Women’s Dead-End Work” was to map the companies currency. Currency is this context means the things and/or tasks that are valued by leadership enough that it can be traded for a higher salary. The currency at one workplace can be dramatically different from another, and you might have to spend a week or two to really identify what it is a your workplace.

Sometimes it can be very straightforward things like how much you sold for - but other times it can be more intangible things closer related to personality traits.

When I identifies the currency at my workplace I also understood where I should redirect my focus.

3 - I tracked my tasks and the added value

Next and also an exercise from the book was to track all the tasks I did and how close or far away from the company currency they were. Shockingly I discovered that, although I did everything and more of what was my job description, almost nothing was closely related to the currency.

This was a real eye-opener. Now I had to find ways to either swap out some of my tasks, or more creatively connect my tasks to the currency. This took some times to figure out how to do, but along the way I managed to not only get rid of a lot of overtime work, but also to successfully connect almost all my tasks to the currency directly or indirectly.

4 - I repositioned myself

Now that I understood how my tasks were connected to my workplace’s currency I could really get creative and start positioning myself in ways where my work would get mentioned and I could more naturally display it and get noticed. In that process I started to talk more about my work in the light of our currency. I stopped saying “This design is good because it helps the user navigate the website“ and instead said “This user journey is good because it increases the numbers of user who complete the checkout flow with X% and helps lift our sales with X amount a year.”

Suddenly my design work was a business neccessity and not just something nice-to-have.

5 - I learned myself to not be a perfectionist at everything

Because I understood what things mattered to the business, and I needed to redirect some attention to that, I realized I also had to either remove work or complete some tasks faster than usual. As so many other designers, I am a perpetual perfectionist. Everything needs to perfect and I would fight to death and argue the importance of every little decision all the time.

But I realized that while the company had a currency, I also had my own. Things that were important and things that weren’t. So I stopped fighting life and death about things that were either not close to the currency or where I didn’t really care so much if I just took a moment to think about it first.

Why did I even fight over things that didn’t really matter to me! So silly to look back at now.

Today I often stop and ask myself how important a task is to me, before I decide how to go about it. And man oh man has it lifted some weight off my shoulders.

Subscribe now

6 - I stoped fighting and started dancing

Now I no longer fought for tasks or decisions that had no value to me nor to the company - and I had stopped fighting to get others to change. And that already lifted my mood on an everyday basis. I realized it had taken a lot of energy out of me to constantly fight.

Instead I had a clear understanding of how to interact with my colleagues and leadership in a more graceful way, that were not only increasing my perceived value but was also a lot more fun and enjoyable to myself and others. Now I could position myself in a way that worked around others regardless of what they did, and where work dynamics were all of the sudden a dance and not a to-the-death boxing match.

7 - I learned how to negotiate

The process of completely changing how I thought about work also changed how I walked into a pay raise negotiation.

First off I now make my work visible every week. And whenever I talk about my work I make sure to connect it to the company value.

Second, I increase selected efforts in the two months up to the negotiation. I make sure to know what leadership is up to so I can connect even more with them and with my work during this time. I make sure to do this with all the people who directly or indirectly have a say in whether I get a pay raise or not. It dissent have to be every day, but you do need to make impressions on everybody at least a few times for them to remember but not get overly suspicious.

I also keep track of what I have been doing since the last time, so I can go in and easily showcase that.

Thirdly and most importantly I go in and I make sure to let them know what new value I will bring going forward. You guessed it, again they are closely connected to the company currency. If they can see that the value I will add over the next 6 or 12 months is significant, that makes it a lot easier for them to agree to a raise, benefits or whatever you are looking for.

I also realized that understanding what motivates the people I negotiate with is crucial. What are their values and motivation, who do they negotiate with themselves and what do they need to prove to someone else? And what do they think is a good bargain?

In some countries negotiation is dreaded, because it feels like one person wins and one person looses. But if instead you change perspective and considers a good negotiation as a win-win situation, then it can be fun, it can develop you and both you and your employer can gain something more, instead of one being a looser and one a winner.

And that is it!

Still unsure how to get started?

When I first started I wish I would have had a framework that could guide me through this - that way I would also have saved 2 years and gotten there much faster.

If you are there right now, the help is near!

I have developed a workbook that helps you step by step through this process with exercises that support each step of the way.

I will take you all the way through defining the company value to negotiation tactics and task tracking as preparation. I also added a few more useful exercises not described here that can help you navigate the office policies and bring out your efforts, work and value without compromising your own values or losing goals in the process.

As a designer and creative myself I know the struggle and I build the workbook as a practical and easy to use tool that I would have wanted myself.

Get the workbook here

Thank you for reading!

There are more ways you can show your support:

❤️ Tapping the heart below helps my content get discovered

📧 Forward this to a friend

🌟 Become a subscriber if you are not already

I know this is a controversial statement, and that many of my peers preach the opposite. They package their best and most tested prompts and sell them. They teach others how to build prompt libraries. And when I talk at seminars, I am often asked about how I organize my prompt library. And my answer is, “I don’t, because I don’t have one.”

Because to me, a prompt library is only working against me.

To have a prompt library requires that it is organized, updated, tested, and then adjusted when one finds better ways. It gets long - and then it needs even more organizing.

But more importantly, I think it is useless.

Because having a library will weaken my sense of crafting good prompts - which, by the way, I don’t think is very difficult to begin with. Maybe it’s because I’ve managed bigger projects over many years that required me to formulate tasks for other people. A good prompt - or might I say brief or project description - requires you to know who you are talking to and what their background and context are. And this is where I think a prompt library has it’s biggest fault.

Because if you keep crafting the same project description for someone the same way over and over again, without considering their current knowledge, capabilities, experience, and context - then the project description or prompt is not good.

Imagine working with a new colleague and continuing to give them the same task details without considering the growth they’ve gone through from the first day to one years later.

Because AI is changing so fast; models are constantly updated, sometimes I use my private ChatGPT and sometimes I use my work ChatGPT - both of which have different knowledge and context about me - The two models would simply complete a task entirely differently.

Besides that, I would miss out on the constant practice of writing instructions, doing the thinking, and improving my own capabilities. And even worse, I would perhaps miss out when model updates come out and new guardrails have been implemented or functionalities have been added or removed. Every time that happens, I would need to test all my prompts again, register what changes it caused, adjust them, and bla bla bla… You see my point!

The fresh blank slate without a predefined prompt allows me more space. Sometimes to make prompts that are more exploratory, with very little details - and to discover how much the model has improved and what new perspectives it returns when it is not given a long, detailed prompt. To be frank, I often start with very few details, just to see what happens and where it might take me. And sometimes that leads me to really interesting new perspectives. Other times it sparks a new creative idea that I can build on. And other times, I simply add more details to my prompt after getting the first initial reply back.

But now you might think, “Well, not everybody has a ton of experience in crafting a clear prompt, and wouldn’t a library tested by someone else help them out?” …and yes, it would. No — actually, that’s a lie. I don’t think it helps them, because they will never get better if they just copy-paste from someone else’s library. They will never understand why that library came to be as it is - what tests and agendas were behind it, what type of person made it, and whether it really gives good answers or just slightly better answers than what you’d get writing your prompt from scratch. They simply don’t know your context, experience, or capabilities well enough.

It’s like making a copy of an artwork and calling yourself a good painter. Unless you start painting, you will never get better at painting. But if you practice over and over again, then eventually you will be able to paint whenever, whatever, and in many different styles for different purposes.

And while you’re learning to paint, I’m pretty sure you won’t use that prompt library one bit.

Do you agree - or perhaps I’m missing a perspective? Let’s discuss in the comments. 🧑‍🎨

Thank you for reading!

There are more ways you can show your support:

❤️ Tapping the heart below helps my content get discovered

📧 Forward this to a friend

🌟 Become a subscriber if you are not already

We probably all do this - I know I am guilty from time to time.

Today we set out to find a solution for one of my colleagues who is particularly prone to questions like that.

But setting up an automation sounds so daunting, right! What system do we use, is the data safe, who has time to help with the technical part, what specific questions are we solving first, and what channels should we include?

A thousand questions as to how to solve the thousand questions. How ironic.

And then we just decided to go simple.

One question to be solved. One channel. One hour. Three people: the person getting all the questions, a developer with a little extra time, and me to translate between the two - and to soak up knowledge, which was just as important.

We decided on a question for which we already had a small video recorded that showed how to do something. We started with Slack and N8N (which is a well-known and reputable AI workflow and automation system) because we already had another flow tested between Slack and N8N. And then we got to work.

The three of us sat down over Teams and made the setup together. It took a few tries, fails, and tests, but after one hour we had a fully working solution ready.

One hour which will be earned back in less than one week.

Not only did we have the solution ready, but my colleague was so invested, excited, and during the small breaks (where our developer was testing), she observed and started to forme a completely different understanding of what AI and automation can do for her. She also wrote down several other ideas for other types of tasks that she was now able to imagine being solved a similar way. When she saw the solution come to live, she got so exited about not having to put down heavy complex data sets, to answer trivial questions (she works as a data analyst). Maybe you can imagine how much this meant for her.

And lastly - because she had seen and followed the developer’s test on the screen - she was now also able to changes and add-ons to the automation flow. Something that would have been unimaginable without the waiting time and observation along the way.

But do you know what the extra, extra, extra bonus of this 1-hour investment was? We now have a framework setup that we can copy to every single colleague. And we have a colleague who ran back to her close peers and, overly excited, told them about this task that was now off her hands!

So if we set a limit of one hour’s work, what task could you find another solution to tomorrow?

I would love to know yours in the comments 👀

Thank you for reading!

There are more ways you can show your support:

❤️ Tapping the heart below helps my content get discovered

📧 Forward this to a friend

🌟 Become a subscriber if you are not already

Subscribe now

AI is changing the role of designers faster than ever before. Some creatives embrace it; others fear it. Both reactions are valid. But as AI takes over execution, designers must evolve from makers to directors - focusing on strategy, taste, and judgment. The future of design isn’t less human; it’s more human, guided by designers who understand how to combine creativity, technology, and context.

As a design consultant and AI specialist, I meet many designers, creatives, business leaders, and managers. And the verdict is clear: design is becoming much more difficult to sell and argue the value of.

Designers often react to AI implementation in two ways — with enthusiasm or with skepticism.

The AI enthusiasts

Some designers started using AI early on, jumping on every new tool they find. They keep a few, but most are discarded because the quality doesn’t meet their standards or the tools make fine-tuning impossible. Think of Midjourney, where a small request for detail can unexpectedly change the entire image.

The AI skeptics

Other designers see AI as an immediate threat to their jobs. Some have experience and can see how good certain results can be and worry their role may soon become obsolete, while others have little to no hands-on experience and see AI as a threat to the design craft, creativity, and originality as the primary concern.

Both perspectives are valid because almost no one can keep up with the speed of AI development. Even I, who work with AI full-time, am reminded of this daily.

But regardless of how good AI becomes, we can never make good design without a designer.

From execution to direction

As designers, we must start adjusting our perspective, role, and identity - from being designers and executors of design toward becoming design instructors, decision-makers, and judges. Our taste, understanding of context and payoffs, and the ability to make recommendations will play an even bigger part in our future design roles. They already play a significant part in many designers’ work today, but they are often overlooked or not understood as part of the design discipline.

When we start to talk about designers in terms closer to strategy, the value becomes more obvious.

This is not an easy task for many designers, though. When you are not used to verbalizing how you create value in relation to a strategy, it requires new skills, practice, and a solid framework - until it becomes second nature and part of how we communicate design’s true impact.

AI cannot replace the creative team

This article is part of a series that explores how designers and creatives can navigate the changes that come with AI. The designer’s identity shift is the first piece and the next couple of articles will explore more about why designer might resist AI, why the value of their work is difficult to communicate and how they can work in sync with the company strategy as lastly, a work kit that step-by-steps help designers through this AI shift.

The human advantage

Today we see firsthand how design jobs are becoming scarcer, design is less sought after, and designers must, to a much higher degree, argue why they still matter. If design was difficult to see the value of before, AI is now just amplifying that with the speed of light.

But human designers still have an essential advantage in context, responsibility and empathy. Because AI cannot give itself instructions based on something that just happened or was discussed at the coffee machine. It cannot own decisions or be accountable for the outcome. AI cannot read between the lines and sense the user’s or the client teams’ emotional state or pick up on subtle preferences or personal cues.

You can, of course, argue that most of the things on the list here can actually be performed and handled by AI. And yes, I agree. But only through human training, setup and supervision and in such a complex system that the system itself would likely evolve before the setup was complete.

Even then, humans will remain better at human-to-human understanding and relationship building. And isn’t that what design is ultimately about - understanding people, building relationships and designing outcomes that help the very same?

Human 3 weeks, AI 3 min

I was recently shown a tool that could build website wireframes in just three minutes, and the wireframes looked a lot better than anything I could have done (I used to work in UX design). Everything was perfectly positioned, every button state was accounted for, and it was insanely impressive to see the program build an entire website in detail in front of our eyes in just three minutes.

The demonstration prompted the question:

“Why do we use so much time in the UX process when AI can do it this fast?”

It’s a fair question - for someone with no UX design experience. But every UX designer knows that wireframes are just the output, and not the work.

“Building wireframes isn’t what takes time. Understanding the client’s business, processes, values, and strategy - and translating that into meaningful and strategic design - that’s what takes time. And that’s something that this tool isn’t solving” I answered.

Designers + AI = ❤️

But that does not mean that designers cannot benefit from AI. They definitely can. But tools are only powerful in skilled hands. A design tool without a designer is like a laptop that no one turns on or like an excel sheet used for designing website. It’s useless.

I am a huge fan of AI, and thrilled by how many of my tedious tasks it has reduced or entirely removed.

Our focus should be to find ways where we can use AI tools as tools. And then we can focus on the human centered part of the design work, where intuition, empathy and strategic thinking makes a difference to the products and to the people who will use and own our designs.

The role of the designer is not disappearing, but it is evolving - just as it and we should.

So what now?

Now we must start practice speaking about design in a different way. We must create a narrative where design is not dead and designers are not doomed, but where designers with AI tools play a key role in the companies strategic plan. One where the designer designs and the tool is a tool.

Thank you for reading!

There are more ways you can show your support:

❤️ Tapping the heart below helps my content get discovered

📧 Forward this to a friend

🌟 Become a subscriber if you are not already

About the author

I work at the intersection of strategy, communication, and technology, helping companies turn AI from an abstract idea into a practical tool that drives real results.

As a Senior System Designer and AI Adoption Specialist, I focus on making AI approachable — not just through strategy decks, but through everyday use.

On this Substack, I share insights, examples, and reflections from real projects: what works, what doesn’t, and how AI can enhance creativity and collaboration without losing the human touch.

I hope you’ll follow along!

Subscribe now