Mike Krieger

The most effective AI product development happens when product teams are embedded in the post-training and research process rather than just building UX on top of finished models.

"The functional unit of work at Anthropic is no longer take the model and then go work with design and product to go ship a product. It's more like we are in the post-training conversations around how..."

Product teams provide unique value in AI through strategy, making complex capabilities comprehensible, and demonstrating the 'art of the possible' to users.

"I think there's still a lot of value in two things. One is making this all comprehensible... Two is... strategy, how we win, where we'll play... And then the third one is opening people's eyes to what..."

Defensibility for AI startups comes from deep vertical market knowledge, specialized go-to-market relationships, or radical new interface form factors.

"I think things that are going to, I can't promise this as a five to 10 year thing, but at least one to three years, things that feel defensible or durable. One is understanding of a particular market...."

Effective prompting involves pushing the model out of its 'polite' default state and using automated tools to optimize prompt structure.

"With Claude sometimes I'm like, 'Be brutal, Claude, roast me. Tell me what's wrong with this strategy.' ... It forces it to be a little bit more critical as well. The last thing I'll say is... watch o..."

Standardizing protocols (like MCP) for context and memory is more scalable than building individual, non-repeatable integrations.

"MCP really tried to tackle that middle one [context and memory]... what if we made this a protocol and what if we made this something that was repeatable? ... I hope that's where we had, and I hope th..."

High-velocity AI engineering teams are shifting from line-by-line human code reviews to AI-driven reviews and human-led acceptance testing.

"The team that works in the most futuristic way is the Claude Code team. They're using Claude Code to build Claude Code in a very self-improving kind of way... they would do very line by line pull requ..."

Executive onboarding requires a delicate balance between observing existing culture and implementing necessary strategic changes.

"I actually hadn't joined a company since my first internship in college basically. And I was like, 'Oh, how do I onboard myself? How do I get myself up to speed? How do I balance making sweeping chang..."

Shutting down a product requires recognizing when the 'energy' isn't in the system and having the intellectual honesty to call it early.

"The confluence of those three things [mobile web deterioration, lack of viral spread, remote work]... we entered I guess 2024 and said, 'Look, there is a company to be built in the space. I'm not sure..."

AI products require a shift from traditional Web 2.0 engagement metrics (like time spent) to metrics that capture utility and task completion.

"Traditional engagement metrics might be misleading when depth matters more than frequency... I think overall though... I think you know when your product is really serving people... so much of when yo..."

Differentiate by leaning into your unique strengths and 'personality' rather than trying to beat a dominant incumbent at their own game.

"How do we figure out what we want to be when we grow up versus what we currently aren't or wish that we were or see other players in the space being... I think there's room for several generationally..."

AI can now act as a collaborative partner in the early stages of product planning, from market analysis to user need definition.

"Can Claude be a partner in figuring out what to build? What the market size is if you want to approach it that way? What the user needs are if you look at a different way? ... Models can do that today..."

When AI removes the coding bottleneck, the new constraints become organizational alignment and the technical infrastructure for merging and deploying code.

"We really rapidly became bottlenecked on other things like our merge queue... We had to completely re-architect it because so much more code was being written and so many more pull requests were being..."