I am a strong believer in AI assistance and I use it every day.
I use it when programming our AirGradient mobile apps for iOS and Android. I use it for data analysis, proposal writing, documentation, and research. Even when I am starting out with larger, less administrative or menial, tasks, like exploring ideas that would have taken much longer to test only a few years ago.
Beyond my personal practices, I encourage our team to push the boundaries of what is possible with AI, because I believe it can make us faster, more capable, and more competitive.
The results of this push are happening already. Our development team is moving faster and our documentation and customer support are improving. Overall, using AI allows us to take on more challenging work with a small team. Across AirGradient, people are finding creative ways to use AI to remove friction and increase quality.
But I also see something that worries me. AI does not just help people work faster. In some cases, it seems to make smart people stop thinking.
Recently, I have seen an AI-written customer email that sounded polished but was factually wrong and just did not sound like the sender’s own voice and language skills. In another instance, an impressive-looking chart created by AI had a mislabeled axis and lines that did not make sense.
I have also seen AI proposing a data-analysis methodology that looked mathematically sophisticated but was not internally consistent, and summarized a legal document while missing important attachments and I have seen AI-generated code where the test cases (that were also AI generated) were not realistic, thus implying a false sense of performance.
In each case, the result was presented to me with confidence. But after I asked a few basic questions, the errors became apparent within minutes. The people who presented these cases are smart, capable colleagues and once we looked closer, they immediately saw the shortcomings themselves. So the uncomfortable question is: why did they not recognize the AI’s mistakes before presenting the work?
I think the answer is automation bias. Research describes this as the tendency to over-rely on automated recommendations. A 2025 review in AI & Society argues that human engagement and independent verification are among the most important ways to reduce misplaced trust in AI systems. Microsoft Research surveyed 319 knowledge workers and found that higher confidence in generative AI was associated with less critical thinking, while higher confidence in one’s own ability was associated with more critical thinking.
This matches what I see in practice. The danger is not that AI makes mistakes, which we are all aware of. The danger is that AI makes mistakes in a fluent, authoritative, beautifully formatted way, giving us the feeling that the hard work has already been done.
That feeling is dangerous.
The strongest evidence I have seen for this comes from the “jagged frontier” research by Harvard, MIT, Wharton, and BCG. In a study of 758 consultants, AI helped people perform significantly better on tasks within its capability frontier. But for a complex task outside that frontier, consultants using AI were 19% less likely to produce correct solutions than those without AI. This is exactly the problem: AI can be excellent for one task and harmful for the next, even when both tasks look similar to the human user.
We must resist the temptation to blindly trust an AI's output, and never turn off our brains! This is why we need some rules about using AI in our worklife.
The Lessons we Should Draw from This
What this means for AirGradient, is that we should use AI aggressively, but not blindly. We should use it to draft, analyze, summarize, code, explore, test, and challenge our thinking, but there should be no-go areas.
Using AI-written emails to communicate with customers should be one of them. Customer communication must be human centered and we should put the effort into taking every single one’s issues personally. It must carry responsibility, empathy, and context. AI can help prepare background information, but the final message must be written by a human. I would rather accept spelling and grammar mistakes in a hand written email over a polished AI email. Authenticity can not, and should not be replaced by AI.
We should not use AI-generated images in our blog posts or on our website. Our work is about trust, air quality, science, health, schools, homes, and communities. Synthetic images may be convenient, but they erode authenticity. We should show real devices, real people, real environments, and real data.
If we use AI for blog content, we must edit it heavily. It must bring a real benefit to the reader. If it does not add clarity, insight, or experience, but is just generic AI text, we should never publish it.
Most importantly, we must keep the human touch. Work with people. Connect with communities. Listen carefully. Stay humble.
AI should not replace judgment. It should provoke judgment.
Before we present AI-assisted work, we need to ask simple questions. Is this factually correct? Did we check the source data? Are the units right? Are the axes labeled correctly? Did the AI read all relevant documents, or only the main file? Does the methodology work in real life? Could we explain this without the AI? Do we actually believe the result? Did we scrutinize and challenge the work from the AI, the same way we would coach a junior?
If the answer is no, the work is not ready.
This is not about slowing down innovation. It is about protecting the very thing that makes innovation valuable: human responsibility.
AI can make a small company like AirGradient much more capable. It can help us compete with larger organizations. It can help us build better products, analyze more data, code faster, improve support, and communicate more clearly. But only if we control it. Only if we understand what the AI is doing. Only if we remain aware of how our use of AI affects our customers, our staff, and the wider community.
The companies that win with AI will not be the ones that blindly automate everything. They will be the ones that combine AI speed with human care, skepticism, and accountability.
So my message to the team is simple: use AI more, not less. But never outsource your judgement.
AI should make us sharper. If it makes us less curious, less careful, or less willing to think, then we are using it wrong.
____
For transparency, this article was AI-assisted. Below is the full prompt that I provided ChatGPT. After spending a good amount of time on the prompt, the ChatGPT output was heavily edited. In the end, we spent around 2 hours fine tuning this post. Many thanks to Emilia and Ethan for contributing to this piece.
Sharing this prompt is also an example of the kind of AI use I want to encourage at AirGradient. I had thoughts I wanted to communicate, but I didn't have the time to spend hours turning them into a polished article. Instead, I invested that time differently: into creating a detailed, carefully considered prompt that captured my real views, and then into editing the output until it actually said what I meant. It's not a human-written article, and it's not an AI-written article either. It's something in between, and I think that's fine, as long as long as a human remains responsible for every claim, and as long as the thinking behind it is yours.
Original Prompt:
Write an opinion piece from Achim Haug, CEO of AirGradient covering the following:
- I am a strong proponent and believer in AI assistance and use it a lot, e.g. programming the mobile airgradient apps for ios and android, using it for data analysis, for proposal writing etc.
- I motivate my team to really push the boundaries of what’s possible with AI, however there are a few no-goes:
- Never use AI written emails to communicate with customers
- Never use AI generated images in our blog post or on our website
- If we use AI for writing blog content, we need to edit it heavily and make sure it does not sound AI and really brings benefits and is interesting for the reader
- Always keep the human touch, work with people, connect with communities, be humble
- I strongly believe that if we use AI correctly, we can hugely benefit from it as a company but we need to control it and always be aware what we do, what the AI is doing and how it impacts our wider community
- It’s great to see that our staff is using AI more and more and doing cool stuff with it. I see our development team getting faster, I can see our documentation and customer support getting better, I see we can do more challenging tasks. There are many positive examples throughout the whole AirGradient team.
- HOWEVER, I also see something that worries me. It seems some of our staff is awestruck by the AI and STOPS THINKING. Recent Examples:
- AI writes an email to a customer which sounds great but is factually incorrect and also does not fit to the verbal language skills of the sender.
- AI helped to analyse a big dataset and created an awesome looking chart, but the chart just didn’t make sense when you looked closer to it (e.g. axis was wrongly labelled etc).
- AI helped to develop a methodology for data analysis that was inherently not consistent and would not have worked in real life, but it created some cool looking formulas.
- AI analysed a large legal document but did not read the attachments and came to some wrong conclusions.
- AI created code for training a machine learning model, but it seemed the training had zero effect but automated tests that the AI showed results that were not realistic (and not human verified)
- In all these cases, the results was presented to me. By carefully looking at the data, asking some questions, within minutes it became obvious that what was presented was not methodologically sounds, or in some cases plain wrong.
- These are all smart people from our staff, and just by asking my questions, it became very quickly obvious to them that there were major short comings in their presentation.
- I wonder why did that happen? I have the feeling that the AI makes a blindly trust the result? Does it make switch off our brain?
- What is the danger of this? Does my push for using AI, actually makes our staff dumber?
For above article, also make some research into that issue. Maybe you find some relevant research that we can quote.
Be quite reflective and opinionated on this piece. I want people to really reflect on this.
In the end this is about how a company should use AI, how we can make sure our employees benefit from it, how we can stay innovative and competitive.
