Produktland

On Monday morning I would go in and ask for access to the codebase.

Frederick Andersen is the founder of EDL, a nine-person design engineering studio in Copenhagen that builds design systems for B2B SaaS. This summer he spoke at Figma's Config in San Francisco, with five colleagues in the audience, about moving his designers into the clients' codebase. Everyone expects AI to push the price of design down. At EDL it went the other way. They stopped selling wireframes and started selling product infrastructure: the design system as production-ready code with guardrails, so everyone in the company can prompt and still get something good out. The average price of a project rose markedly. Frederick also sees the middle disappearing in his own field: screen by screen in Figma matters less, while UX research at one end and the structure of the design system at the other matter more. The designer's value is moving out to the edges, and the way there goes through the codebase.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why EDL stopped selling wireframes and started selling product infrastructure
• What a design system is when it lives in the code
• Guardrails: how to constrain an AI so everyone in the organisation can prompt to production
• Design engineering as a role, and why the combination of taste and code used to be so rare
• A month of work, a two-hour handoff: why users end up with something other than what was refined
• The conversation with developers that works, and the one that does not
• Why the average project price went up when they started delivering code
• The floor has been raised, and the ceiling has been raised with it
• The middle disappears: UX research and design systems become the edge

You cannot prosecute an AI.

Rune Lippert is SVP of Customer Platforms at Bankdata, which delivers IT to a group of Danish banks. He came from JP/Politikens Hus at the end of 2020 and has almost six years behind him with digital channels, customer relations and core systems. It is critical infrastructure, and that changes the maths of how fast you can let AI loose. According to a Gartner study, reading and writing code takes up 14 percent of a developer's day, so the whole AI debate is about the smallest part of the work. When AI writes the code faster, the bottleneck moves to review. At Bankdata a human is always accountable for code going into production. Rune's answer is to put agents on the validation, and then quality can end up higher than before: no human ever managed to check every change against security policy, architecture, documentation and the system handbook. AI moves the work up, from producing to standing behind it.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• The Gartner figure: reading and writing code is 14 percent of a developer's day
• Why review becomes the new bottleneck when code is written faster than anyone can read it
• Agents that validate against security policy, architecture and documentation
• Why leadership gets overtaken by its own teams, and why that is a healthy sign
• The design system that is open to AI, and the prototype built overnight
• Hackathons, AI guilds and the phase where a hundred ways have to become one
• The arms race: attacks get cheaper, so defence has to rise with the same tools
• Why Bankdata is looking for product developers rather than UX designers and developers
• The pitfall when every slide deck suddenly grows 75 percent

AI gets you to the hard part faster, but the hard part is still yours.

Tim Herbig has spent fifteen years in product management. He became a Scrum product owner by accident in 2010, went on to be product lead at XING, Germany's LinkedIn, and since 2019 he has coached product organisations around the world full time. He wrote Real Progress, the book about connecting product strategy, OKRs and discovery into one system. This is the first episode of Produktland in English, so the whole conversation is in English. The efficiency conversation is the boring half of AI. The interesting half is what is left to differentiate on, now that eighty to ninety percent of AI features run on the same handful of models. Technology was never a real differentiator, and it differentiates less than ever. Shipping ten times more features solves nothing in a company whose customers never adopted the features they already had. And AI has opened a whole new world of alibi progress: vibe-coded prototypes, OKRs drafted by Claude and generated strategies that feel like work without changing anything.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why technology was never a sustainable differentiator, and why AI makes that obvious
• Shipping 10x more features when customers never adopted the old ones
• Ravi Mehta's line: AI allows you to get to the hard part faster
• Alibi progress in the AI age: prototypes and OKR drafts that feel like work
• The Damocles sword: efficiency from the top, fear of being cut on the ground
• Token usage, output per employee or bottom line: which metric proves the AI bet?
• Why a domain expert with AI outruns a product manager with access to code
• Slack versus Teams: you can't out-product distribution
• Inside Tim's own setup: an Obsidian vault, Codex and a Qigong coach transcribed with Wispr Flow

AI can build anything, but it cannot tell you what is worth building.

Luciano Lykkebo decided in the middle of the corona lockdown that if we all survived, he would build his own conference. He had no background in product design. UX Nordic was meant to gather a hundred people and ended up with almost five hundred. Today the conference is called Future Product Days, it fills Lokomotivværkstedet in Copenhagen with thousands of product people, and Don Norman takes the stage twice. That conference has dropped its AI track. Not because AI matters less, but because it is everywhere, the way mobile once became part of all product development. And when Luciano talks to his speakers, one word keeps coming back: purpose. AI will soon build and answer almost anything, but it has no job to lose and no need to live in this world. So it cannot tell you what people need.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• The conference born during corona: a hundred hoped-for attendees became five hundred
• Why the AI track was dropped: AI is part of the product, the way mobile became
• The AI strategy trap: users meet your product through AI before your product gets its own
• Purpose as the recurring theme among speakers: AI does not know human needs
• Seven or eight people run a huge conference with AI in the engine room
• The programme goes out as a markdown file, so attendees can ask their own AI
• Design engineers, vibe-coding designers and education as a subscription
• From MS-DOS to Windows: a Claude for consumers is coming
• Monday advice: try using your product through the AI, with no interface at all

If your calendar looks the same six months from now, AI has not changed anything.

Ida Wolffram is head of product development for electricity and gas at Norlys, and she started out as a UX designer with Axure and Sketch. Today she sits in an industry moving from delivering power, full stop, to being a tech company with data as its raw material. Everyone says they use AI, and they do. Most have simply put a new engine in the car and are driving the same route. Mail, slides and Copilot is adoption, and Norlys is past that phase. Transformation only begins when the processes, the organisation and the division of labour change, and it does not start with a hunt for use cases. It starts with finding waiting time, repetition, handovers and bottlenecks.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Drop the hunt for use cases: look for waiting time, repetition, handovers and bottlenecks
• AI as RPA on dope: it does not just automate the process, it gives feedback on it
• The two teams at Norlys working their way through the entire product portfolio
• Prototypes so finished that developers no longer wait for the backlog
• The Djøf survey: eight in ten managers use AI, only one in four feels equipped
• The danger of the lightning-fast analysis phase: you fall in love with your own idea and forget the customer
• Why AI makes the UX mistakes we made ten years ago
• Ida's second brain: a strategy that lives dynamically, and a sparring partner set up as her best opponent
• When your agents and mine are just mailing each other: where should humans be good in five years?

You are allowed to be angry that an artist's life work has been scraped, and use the machine at the same time.

Esben Kjær Ravn has run the game studio Kong Orange for fifteen years and directed Patrick Stewart in Felix the Reaper. His industry gets punished for something it cannot prove it did not do: Steam only requires a declaration, but users review bomb, and right now what he calls the scent of AI around a game is enough. Meanwhile there is AI in Photoshop, and the answer at the top of your Google search is generated. That is the whole episode in one sentence: code is easy for the player to swallow, the art is not. The rest is about what you do when you neither want to lie nor stop using the tool.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why a 3D model or a voice is far more sensitive than code: the machine is trained on real people's art
• Steam does not punish, but the loudest users do, and a shitstorm lasts weeks
• The synthesizer: music will never be the same, they said back then too
• Where the industry actually dares to use it: the early idea phase, rapid prototyping and market fit
• The agents that could have ported Felix the Reaper without costing the programmer his job
• Rockstar's statement about never using AI creatively, and why Esben thinks it was forged by lawyers
• The danger of educating a whole generation into an illusion of an AI-free production environment

You are an editor. Your professionalism is about your taste.

Tav Klitgaard is group CEO of Zetland and has spent 11 years building what he calls a human product. Now he faces a paradox: Zetland uses AI everywhere it possibly can, precisely to make the product as human as possible. The programmers are mostly editors now, and the word has spread from the newsroom into the whole company. That is the whole episode in one sentence: code is free, the article is headed the same way, and what remains is the ability to choose right among a thousand suggestions. That muscle needs training.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why Zetland's programmers are mostly editors now, and how the word has spread through the whole company
• The paradox: maximum use of AI to make the most human product
• The principle "no published word from an AI", and why it is becoming a straitjacket
• The thought experiment with two nearly identical audio files, one of them AI, and why the real one is worth more
• Why the product is not the article, but what the article does to you
• Commodity journalism and the business model after Google
• Good Tape: from internal newsroom tool to standalone spinoff
• Build vs. buy: "we'll just build it ourselves", and the energy bill that follows

Agentic, autonomous, secure. Pick two.

Rasmus Dalsgaard Andersen hacks companies for a living. He is just there before the real hackers are. His best story is about himself: he built a frontend and a backend with Claude Code, and it worked. The day after, he discovered there was no authentication between them, and that the backend was wide open to the internet. The model had not mentioned it. He had not asked. That is the whole episode in one sentence: the model does exactly what you ask for, and no more. The rest is about what you then do about it, when you are not a security guy yourself.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why 45 findings after an hour of coding is a problem, even with the best models
• Agentic, autonomous, secure: why you can only get two of the three right now
• The compliance agent that blocks his own deploy when data is headed for the wrong country
• Background loops that scan your tech stack, so security leaves your head and moves into an environment that is always on
• What actually topples companies: not zero-days, but misconfigurations and bad passwords
• Shadow board: the CTO, the CISO and the procurement officer as agents that shoot your product down before the customer does
• The EU AI Act as GDPR for AI use, and what it means for what you are building

Users do not need a lot of things. They need a solution to their one problem.

Peter Gregersen has done UX design for almost 20 years and calls himself a grumpy old designer. We talked about what AI actually changes in the design craft, and what stubbornly stays the same. His point is easy to miss in the AI hype: users do not need a lot of things. They need a solution to their one problem. The tools get faster and cheaper, but that does not make design skill less important. It makes it more important, because now everyone can build, and someone still has to answer what the purpose is.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why the discipline does not change, but the way of working clearly does
• The bank bot that gives you a sourdough recipe while you check your account: meaningless use of AI
• AI slop: what it is, why it happens, and why you should stop making it
• From pushing pixels in Figma to vibe-coding in the production pipeline at Vipps MobilePay
• The chat interface as a friend who does not understand much: you have to over-explain everything from scratch
• From roles to responsibility: when AI produces the artifacts, who does the quality assurance?
• Why research cannot be replaced by a three-line AI summary, the empathy goes out the window

When time is the only barrier, value goes to zero.

I first met Bo Møller 15 years ago, back when he ran the web agency HTML24. Since then he has built and bought SaaS companies for 20 years: resOS, Alunta, idguard, Boligforeningsweb. He hosts SaaS Købmænd and chairs the board of Morningscore. His point is uncomfortable for anyone who makes a living selling software: execution is becoming a commodity. What remains is the banal and the hard part, solving a real problem for someone willing to pay for it.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why the value of anything everyone can build goes to zero
• How Boligforeningsweb outcompetes players who have owned the market for years
• Bo's setup: ten agents in Conductor, where Codex and Opus review each other
• Tonny: the agent that lives in Slack with its own email and works while everyone else sleeps
• "Stop sharpening the axe, go fell some trees instead"
• What large organizations get wrong that a small operator gets to skip
• Traceability: everyone should be able to trace their work to a paying customer

It is the sewer contractors who have the best products.

Troels Abrahamsen found major success as the frontman of Veto, right as the music industry went from physical distribution to streaming. He has built software all along the way, today among other things Yap, an open source context layer you own and host yourself. His thesis: what hit music first, when distribution became free and the market drowned in content, is hitting software now.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• The music industry as the product world's canary: what happens when distribution becomes free
• The DDoS effect on the distribution layer when everyone can generate 40,000 songs in a day
• Technical promise vs marketing promise: "show me the engine room where the empowerment happens"
• Scaling as the problem itself: when products stop getting people in and start keeping people in
• Yap: a self-hosted, open source context layer that breaks with vendor lock-in via MCP
• Chat as a new kind of browser, and what that means for how you build products

When you can build everything, you should not build everything.

Christian Pedersen is Head of Product at WasteHero, an Aarhus scale-up of about 40 people building software for the waste industry. Most product leaders speculate about AI transformation. Christian is in the middle of one. His vision is concrete: one day, to sit with a customer, hear what hurts, and have a loop running that builds the solution and ships it to production while the conversation is still going. And his honesty is just as concrete: we do not quite know what we are doing. We explore the whole chain and find out where it breaks.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why this is a transformation project, not a tool project, and why you cannot hire your way out of it
• The first place the chain breaks: product documentation, and how WasteHero solved it with markdown in GitHub
• Why he ships greenfield apps straight to production, and still kills 9 out of 10
• The prototype is dead: what is a functional prototype for, if the receiver is an engineer and not a user?
• The product engineer type with a foot in each camp, and why product people need to understand GitHub, branches and pull requests
• Why you should take tasks away from people instead of stacking AI on top of everything they already do

The technology is the easy part. The rest is organization.

Benjamin has worked with AI for over 30 years and has served as CIO and CDO at some of Denmark's largest companies. Today he advises on AI transformation through Munk AI. His diagnosis is not that Danish companies are gripped by hype. The opposite: ambitions are too small, and leadership lacks the imagination to see what is already possible.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why the problem is not hype but a lack of imagination in leadership
• 84% say AI strategy is important, 5% have one, and where it goes wrong
• Personal AI vs structural AI: why Copilot is not a competitive advantage
• AI as a feature is overrated, AI as a change to the user's workflow is underrated
• The rift between stable IT operations and AI speed: where vibe-coding projects run aground

Execution has become free. What remains is taste, critical judgment and responsibility.

Mikkel Malmberg has been programming since he was 12. Now he puts robots to work doing it. We talked about what really changes when delivery becomes free, and what is left to be good at. His point cuts through the hype: it is easier than ever to build software, but that does not make it easier to build something good. Execution is a commodity now; what counts is taste, critical judgment and responsibility.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why most of programming was never about writing the code, and what disappears as a result
• AI slop: the best or the worst thing that has happened to the craft? (The answer is yes)
• Why AI always picks the safe, average design choices, and how to avoid looking like everyone else
• What happens to open source when a PR can be pure prompt, and why projects are closing for contributions
• Why you should send your developers a prototype, not a "finished" solution

You are quickly 90% done, and then all you are missing is the last 90%.

Bjarne designed spaceships for LEGO in the 80s and made Lego Boost with me ten years ago. Today he comes up with names for companies, a job an LLM should in principle have taken over long ago. It just has not: AI can generate, but it cannot reduce, and it cannot feel.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• The 90% problem: you are quickly 90% done, and then all you are missing is the last 90%
• Why AI always adds, while the hard part of product work is cutting away
• Lego Boost today: hard fun, and why voice control would have ruined the product
• Point your AI tokens at discovery instead of delivery
• Vibecoding waste and the 3D printer parallel: apps nobody uses

Roadmaps were designed to manage slow engineering. That era is over.

Dennis was the first product manager at Falcon.io. Now he is building Propane, an AI product for product teams. His point is simple: we have become good at building fast. We have not become better at deciding what to build.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why Dennis thinks roadmaps are dying, and what should replace them
• The large Danish company that wanted to rebuild its entire product stack in six months (and found out it would take years)
• What happens when everyone can build but nobody knows what to build
• Why product sense, not product management, is the skill every organization lacks
• Why you cannot know what AI tools can do for you until you have tried them yourself

The new bottleneck is not code. It is how fast you make decisions.

Ulrik has spent 15 years helping organizations transform. Everyone talks about AI clearing the backlog. Ulrik flips it: if your team runs out of tasks, the problem is not solved. It is exposed. A feature factory just gets faster with AI.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Democratisation of product development: what happens when the ability to build moves all the way into the business
• The new bottleneck: not code, but how fast the organization can make decisions
• Governance as an AI task: an agent that checks your decisions against the organization's principles
• The glitter-paper problem: why polished AI output makes us drop our critical guard

When delivery is free, discovery is the only thing that counts.

Kasper Juunge teaches software developers agentic engineering: AI agents on real production codebases, not prototypes. I talked to him about what happens to product management when code stops being the bottleneck. His point is blunt: the PM job is not under threat. It matters more than ever, because when delivery is free, discovery is the only thing that counts.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• When agentic engineering actually started working on real codebases, and what changed in August/September 2024
• Quality assurance as the real bottleneck: good old software best practice matters more than ever
• The trap for juniors: outsourcing understanding, and what happens to the codebase in the long run
• The biggest mistake Kasper sees in the wild: building things you do not understand, and losing respect for each other's craft

The biggest risk is not that AI takes your job. It is that we build feature factory 2.0 and produce twice as much of the wrong thing.

Natalia spent two years defining product roles razor-sharp at Novo Nordisk. Then AI came thundering in. I talked to her about what actually changes, and what does not.

Open in Spotify → Listen on Apple Podcasts → Watch on YouTube →

Among the things we get into

• Why "hurry slowly" is her most important advice for product people right now
• Feature factory 2.0: what happens when delivery becomes free and discovery becomes the only thing that matters
• The difference between an efficiency task and a qualification task
• What happens to roles when everyone can build everything: from "who does what" to "what do I quality-assure"
• Why regulation in a pharma company paradoxically gives you more freedom to experiment