How I Use AI – with Daniel Gebler 🛒
How Picnic’s CTO is using AI to scale 300 engineers into 1K, which AI tools he uses daily & why he bans AI once a week
👋 Hey, Dan here!
Welcome to this week's edition of In Founders Words, where I interview founders building big from Europe – uncovering their experiences, with actionable advice and practical insights. In their words.
This is the 3rd edition of my series "How I Use AI" – where I go inside the workflows of founders who are putting AI to work day-to-day – to save time, scale teams and stay human.
👤 This week: Daniel Gebler, Co-founder & CTO of Picnic
Daniel Gebler is co-founder and CTO of Picnic, the fast-growing online supermarket that's reimagining grocery delivery across Europe.
Founded in 2015, Picnic has expanded from the Netherlands to Germany and France – bringing their iconic electric vehicles and tech-optimised delivery to millions of households.
As CTO, Daniel leads a team of 300 developers building the tech that powers everything – from the customer app to warehouse automation.
His no-nonsense approach to AI cuts through hype to focus on real productivity gains.
💡 What you'll learn:
How Daniel turns Picnic’s 300 developers into 1,000
Tools and how he uses AI daily
Why AI productivity gains aren't always showing in metrics
Why he mandates "AI-free days" at Picnic
Let's dive in. 👇
📈 How AI Turns Picnic’s 300 Developers into 1,000
Daniel thinks about AI's impact on engineering productivity through a powerful mental model:
"We have 300 developers. These AI tools have the potential to make us twice as productive. So you end up at a capacity of approximately 500-600 developers."
But that's just the beginning. The ultimate boost:
"The next change for productivity (that we are in the midst of now) is the agent revolution, where each human developer gets complemented by AI agents. Nobody knows exactly the impact yet, but conservatively we can expect another 2x boost.”
This means:
Actual number of developers at Picnic = 300
Developer capacity using AI-assisted coding tools = 500+
Capacity with both AI-assisted coding tools and AI agents = 1000
"Long-term, AI-assisted coding, AI engineering and AI agents together boost the development capacity by 5-10x (and may have the potential to go way beyond)"
While acknowledging this is "a bit speculative," Daniel uses this framework to reason backward when approaching new opportunities:
"We are always asking ourselves: what can be done 90% with an agent? What can be done at least 60% with an AI engineering support tool? Such that our old way of working is the remaining few percent."
He believes this future is coming quickly:
"I think by 2026-27 we will get there."
🧰 AI Tools Daniel Uses Daily
Daniel breaks down his AI usage into three distinct categories:
1. LLMs as Junior Analysts
Using ChatGPT and Perplexity for research, best practices, and understanding how others solve technical problems.
"There is the classic LLM for everything that is not engineering: ChatGPT, Perplexity, etc., which I use as a kind of a junior analyst. Although not even so junior, since they also have research mode."
2. Engineering Support Tools
Employing coding tools like GitHub Copilot to support engineering.
"The second category is any form of engineering support. This is specification, implementation, deployment, bug fixing and operations where we're using an already broad set of tools, obviously GitHub Copilot, then also Cursor AI, IntelliJ and Lovable."
Daniel describes how Picnic’s approach to these tools evolved through the hype cycle:
"We went through literally the Gartner hype cycle with these tools. We were very excited in the beginning about what they could do, but very quickly got disappointed."
After initial disappointment in 2022-2023, they've now found genuine value:
"We see this as a kind of co-developer now."
Daniel and his team have been exploring “vibe-coding” platform Lovable, but with careful implementation:
"We have not used it yet for production code, so we have used it for internal prototyping and to get experience of how to set up a proper development process around such a tool."
Daniel remains cautiously optimistic but realistic about how these tools fit into enterprise-scale development:
"If you talk really large scale applications, it's hard to estimate whether these tools can get to that level or not, or if it will only be a little bit... like with low code."
Fun side note: Here’s something I vibe-coded with Lovable last weekend: Dad’s Night Out
3. Thinking & Analysis Support
Leveraging AI to probe specifications, identify edge cases, and to perform qualitative analysis similar to traditional statistical sensitivity analysis.
"The third part is actually not so much pure software development support tooling, but the thinking support side... what I do for instance for a specification, for documentation, or for a new proposal."
This third category is more nuanced – it’s about using AI for exploratory thinking. Daniel uses AI to probe specifications with questions like:
"What are the unhappy flows? What are the edge cases we have not considered if you would do this kind of implementation in 10 years time? How would it look and work differently? How would people have done it 10 years ago?"
Daniel finds this particularly valuable for qualitative analysis:
"What people have done in quantitative analysis with sensitivity analysis, which is a very standard tool that people have used in statistics over the last 100 years, can now also be done in qualitative textual descriptions, which is fantastic."
Daniel uses these tools to guide decisions, accelerate projects and keep his team aligned.
👀 A Day in Daniel's AI-Powered Workflow
If you looked over Daniel's shoulder during a typical workday:
"You would mainly see ChatGPT and Perplexity. You’d also see me checking the AI Slack tool we built ourselves. Then, once or twice per week, you would see a Cursor AI and IntelliJ"
Daniel uses these tools throughout his day to enhance his thinking and productivity:
Morning research: Using ChatGPT and Perplexity to understand industry best practices
Specification review: Probing new proposals for edge cases and extensibility
Weekly summary creation: Generating first drafts of his team updates
The Custom AI Slack Bot 📋
One of Daniel's favourite AI tools is a custom Slack app he uses regularly:
"What I'm using very regularly is a small Slack app that we built for summarisation of all the pull requests that have happened on a day or week."
This gives him a quick overview of his engineering team's work:
"Basically it’s just an overview that summarises what has been changed and what will it change. And how big is the effect of the change. It's a fantastic summary where in 30 seconds you get an overview of what 300 engineers are doing. That, I have to say, it is very strong."
🔬 Three Ways AI Changes the Development Process
When I asked Daniel about how AI tools have improved productivity, he offered a framework,"if you use any form of productivity improvement tool like AI, then three things can conceptually happen":
1. Same Quality, More Output
"With the same quality, you get more stuff done. So not better, but at least you get more done. That’s nice."
2. Same Scope, Better Quality
"You get maybe the same scope done, but with better quality. Also nice. But if your quality before was already good, maybe you didn't need higher quality."
3. Same Results, Less Time
"Maybe you get roughly the same scope in the same quality in less time. And then the question is, are you as an organisation able to reinvest this available time in any meaningful way?"
Here's where Daniel sees a potential trap – Parkinson's Law, where work expands to fill the available time:
"If you use your old planning methods and forecasting methods (based on not using AI tools), then even if a technology would allow you to get it done in half the time, you will still somehow use the entire time and you will not see a productivity improvement."
This insight perhaps helps explain why, despite all the AI tools they use, Picnic hasn't seen the massive productivity jumps they expected in their metrics:
"We have not seen that kind of huge step up that others have been seeing. While we obviously have the perception that we get more done, why are we not seeing that? I don't have a perfect number currently, but this is definitely something I'm a little bit concerned about."
🧪 AI in Action: A Real Example
I asked Daniel to walk through a specific project where AI made a difference.
He shared how Picnic needed to extend their app to allow business customers to order, requiring a complete rethinking of their user authentication model:
Step 1: Background Research with AI
"Step one was researching how, for instance, Apple and Spotify organise family accounts, and how are bigger platforms, marketplaces like Amazon and Bol.com organising B2B accounts."
"I used Perplexity and ChatGPT to research what a pretty good initial version would look like."
Step 2: Implementation with AI Coding Tools
"The team used a mix between three different tools to build this: Cursor AI, and an AI plugin for IntelliJ."
With these tools, they:
Translated the specification into code
Auto-generated tests
Intentionally didn't review the code itself at first
Deployed internally
Step 3: The Reality Check
"Interestingly, it didn't work properly. So we learned, okay, that was maybe a little bit too much trust in the code that was auto-generated."
But the team discovered something important – AI-generated code has distinct characteristics:
"We were positively surprised that the code is actually well written and pretty easy to review - it follows a very clear structure: ChatGPT follows a bit of a structured way of formulating text, and the code is also, in a similar way, well articulated."
The only drawback?
"It is very talkative. So it generates a lot of code, and you can easily write it shorter and faster. But that doesn't matter so much."
👥 AI's Impact on Team Roles and Hiring
The rise of AI is changing how Daniel thinks about team composition and hiring:
"The profile of a developer, what engineers do, how they work, what is the requirement for a good developer, will change quite a bit."
AI for Hiring
Picnic have implemented an AI-driven approach to hiring:
"For hiring, we use an ideal candidate CV that we have built over time, and we basically match new CVs that come in with that."
He acknowledges a limitation of this approach:
"What we risk optimising for is finding people that can write proper CVs, not the best candidates, which is a big difference."
But this AI system helps identify points of interest and discussion in an initial screening:
"In essence, you have some red flags that you want to identify early, and points of attention for the first phone interview. You get basically a call script, which is really powerful."
Will AI Replace Developers?
Daniel he doesn't see the end of developers coming any time soon:
"There's obviously been a lot of discussions about that - will we need developers at all? Or will we need far fewer developers? I can't see it. This is a discussion that has happened in the tech industry for many, many years."
"Every kind of wave, when you went from assembly to interpretive questions to object-oriented questions. When you moved into microservices, when you moved into Agile... Everybody was saying, 'Well, with this new way of working, you need less now developers’ have never materialised."
Instead, he sees a shift in what makes a developer valuable:
"Conceptual and analytical, let's say high conceptualisation, abstraction and analytical skills, which have always been relevant, will become even more important."
Communication skills are also becoming more critical:
"Developers are usually not great at communication, but I think that is becoming an even bigger issue, if to some extent, through prompts, because English becomes at least part of your programming language."
"In the past, we could always get away with, well, you have a bit of awkward documentation, and developers are a bit in their own world... But now we are changing the interface between machines and humans by actually making it use human language."
⚠️ The AI-Free Day
In an interesting counterpoint to his AI enthusiasm, Daniel has instituted something unusual at Picnic:
"I'm encouraging everybody to have an AI-free day... where you simply continue to work in the old way."
While many are focused on governance and security concerns with AI, Daniel believes these are "relatively straightforward to manage." The deeper issue is about preserving core abilities:
"Really learning software craftsmanship, which we all have learned in a specific way – new graduates that come from university that work only with AI tools will no longer have that learning."
But his main concern:
"What I'm really concerned about is, can you still properly review code if you never have learned to write that kind of code from scratch?"
He acknowledges the tension in this position:
"The question is, am I speaking like the old guy that just can't adapt to a new way of working, or is it a valid concern that we need to be really considering? I don't know, but I want to figure it out."
🔮 What's Next? What Daniel is Watching in AI
When it comes to emerging AI capabilities, Daniel is particularly focused on Model Context Protocl (MCP) that allows multiple AI agents to work together on complex tasks:
"What I'm very interested in, and where I think we will see much more, is the transformative power of having a proper MCPs set up so that you can work in a structured way with AI agents, either internally or externally."
He also sees a shift in what becomes valuable in an AI-saturated world:
"Authenticity and security will become an even bigger kind of asset, in a world where content and value can be so easily created by machines."
This extends to personalisation:
"We have tried, for instance, creating product images based on the shopping history for every customer... so I can customise an image of pasta, based on what they have ordered in the past. So a tailored image to what you will actually cook with it, because we can derive pretty well the kind of meals that you cook based on your purchases."
The power of this capability comes with ethical considerations:
"It's also actually super scary what you can derive out of that data. And I think finding the right boundaries for using that kind of technology and not being too invasive is a bit of a concern."
⚡️ Final Thoughts
Daniel's approach to AI at Picnic balances enthusiasm with pragmatism. His productivity model provides an interesting framework for thinking about AI's potential impact, while his insistence on AI-free days acknowledges what we might lose along the way.
Perhaps that's the most important insight – clear-eyed adoption of AI requires both excitement about its possibilities as well as thoughtful consideration.
Thanks for reading In Founders Words. This is the 3rd in my series exploring how founders actually use AI in the real world.
Know a founder I should talk to next? Hit reply or drop me a note.
Until next time,
Dan