Jonas Brandt
“My users are all engineers. Their bug reports are pull requests. It's the best PM job I've had.”
- Martin Knapic, Product MoatJonas Brandt runs platform product at Kestrel Systems in Berlin, which means his users are all engineers and his bug reports arrive as pull requests. Eleven years in, he's become one of the clearest voices I know on the specific ways AI has changed what a platform even is.
Identity & Background
- Name
- Jonas Brandt
- Current role
- Principal Product Manager, Platform
- Current company
- Kestrel Systems
- Location
- Berlin, Germany
- Years in product
- 11 years
Your story: how did you end up in product?
I was a backend engineer who kept getting pulled into 'how should this API actually work' conversations instead of just implementing the spec handed to me. Eventually a lead suggested I just own that conversation formally, and platform PM was the closest title.
Core PM Philosophy
How do you define product success?
Time to first successful call for a brand-new integrator. It's a better proxy for documentation and API design quality than almost anything in a satisfaction survey.
What's your decision-making philosophy?
Flexibility versus opinionated defaults. Every escape hatch you add for one team becomes a compatibility promise you're keeping for years, often for a use case you never fully understood.
How do you manage stakeholders?
Internal developers are my customers and my stakeholders at the same time, which is awkward. I hold a quarterly open review where any team can challenge the roadmap, and I publish what we said no to and why.
Product & Decision-Making
How do you prioritize?
By blast radius: how many teams break if we get this wrong, and how many are blocked without it. Reliability and migration paths outrank features until the platform is boring in the best sense.
How do you do customer discovery and validation?
API usage logs are my interview transcripts. What people retry, what they wrap in a helper function, what they route around - that tells me more than a survey ever could.
What's your favorite product you've shipped?
A deprecation tool that scanned dependent services and opened migration pull requests automatically. It turned a dreaded quarter-long cleanup into a two-week non-event.
AI & the Future
How is AI changing product management?
We now ship inference infrastructure the same way we used to ship storage - as a primitive, not a feature. Teams expect to compose it, not just consume a fixed endpoint.
What's one thing about AI everyone gets wrong?
That platform teams should chase AI features for their own sake. The real opportunity is making the boring parts - logs, incident triage, config review - legible enough that an assistant can help safely.
Beliefs & Opinions
What's your unpopular product opinion?
Cost observability for AI-driven workflows specifically. Teams instrument latency and error rate obsessively and then get blindsided by a token-cost bill three times normal because nobody built a cost dashboard for the orchestration layer. I think that becomes standard platform tooling within two years, out of necessity.
What makes a great product?
One you forget is there. If the platform is doing its job, product teams stop thinking about it and start thinking about their customers.
Company & Product
What does your product solve?
We're a workflow-orchestration platform - the infrastructure other engineering teams use to chain together long-running, multi-step processes across services without hand-rolling retry logic and state machines themselves.
What's the market opportunity?
Against the bigger open-source and managed orchestration platforms, we're not trying to out-feature them - we win on operational simplicity for a team that doesn't want to run and tune the orchestration layer themselves. Managed and boring is the pitch.
Why would you recommend it to a peer?
If you're a growing engineering org whose product teams keep rebuilding the same plumbing, talk to us before your third internal 'shared library' project. It's cheaper to compare notes early.
More from the conversation
What do you do outside of work?
I build mechanical keyboards from individual switches up, which is a strange, deep rabbit hole of a hobby. It's the same instinct as platform work, honestly - caring intensely about an interface almost nobody else will consciously notice unless it's wrong.
Was there a moment you almost left engineering-adjacent work entirely?
I did a year in a pure infrastructure ops role with no product input at all, just running what others designed. I moved back toward product because I missed shaping the 'why,' not just executing the 'how.'
How big is the team, and who's the customer?
Platform product and engineering together are about 60 people, inside a company of 500. Our ICP is an engineering team at a mid-to-large company already running enough microservices that manual orchestration has become genuinely painful.
How has AI changed your own day-to-day process as a PM?
Spec review now includes a section on model-call cost and latency budgets for any workflow step that calls an LLM, which didn't exist as a category eighteen months ago. It's become as standard a line item as error handling.
Enjoyed this conversation?
A new interview with a product person lands every week. Get it in your inbox.