@abreunotes closed with: How do scarcity and abundance shape memorable products?
Mike’s takeaway: you can build almost anything. The constraint is where you put your time, tokens, and team.
Full interview (Portuguese; English subtitles):
youtube.com/watch?v=bDYx8aug…
5/ Building gets cheaper. Serving users still costs.
Productivity AI can justify its price through measurable gains. Consumer AI has a harder equation: tokens cost money every time.
The bet: smaller models handle most tasks; frontier models check the judgment.
4/ The first 90% is getting easy. The last 10% is where companies get built.
Anyone can generate a decent app. Polished workflows and industry expertise still take work.
For Brazil’s AI founders: close the gap between what models can do and what companies actually use.
3/ Agents won’t use your product the way you expect.
Asked to make a part in AutoCAD, Claude wrote the underlying file directly. Taught Google Calendar’s UI, it chose MCP instead.
Study those paths. Build around how agents actually work, not just how humans do.
2/ AI gets more useful when knowledge is shared.
80–90% of employee interactions with Claude at Anthropic happen in Slack. It sees how teams work and “dreams” overnight to learn from their processes.
Private chat histories leave that organizational learning untapped.
1/ Human code review is becoming the exception.
At Anthropic, multiple Claudes hunt for bugs, assess severity, and send findings back to the Claude that wrote the code.
Humans focus on architecture, security, and product judgment. Not every complaint deserves a fix.
In January, humans reviewed 80-90% of pull requests at Anthropic. Today that's only 5–10%.
We hosted @mikeyk for Atlantico Tech Talks. Here are our 5 practical takeaways on how Anthropic builds with AI.
Full interview video at the end of the post. 🧵
Brazil was early for Cognition: Nubank became its first enterprise customer. Since then, Devin has made its way into 10+ of Brazil’s leading companies, including Itaú, Santander, and Natura, helping engineering teams adopt AI at scale.
And now Cognition is doubling down by opening up an office in Brazil and building a local Forward Deployed Engineering team. Perhaps the first major Silicon Valley AI company to make that kind of commitment to Brazil.
Silicon Valley’s best increasingly sees Brazil not just as a market to sell into, but as a market worth building from. It's still early, but we're excited to see what's ahead.
When @AtlanticoPart's Ana Martins invited Cognition's @ScottWu46 to come down to Brazil, we weren’t entirely sure he’d take us up on it.
A Brazil office launch and key customer visits helped align the stars and make this week in São Paulo a reality.
Ana had the pleasure of interviewing Scott today, and I loved hearing him go back to the very early days of Devin in this clip and the massive evolution since then.
Why do @Cognition and other leading AI companies have their eyes on Brazil? more below on what we learned from Scott and the team👇
What a great final episode for @DouthatNYT on Interesting Times. In world where polarization and algorithms lead us to be less exposed (and open) to “the other side’s” view, Ross did a spectacular job at trying to bridge this gap and further this perspective. He will be missed!
Loved the philosophical angles of this conversation on longevity (or “eternal-evity”) with @bryan_johnson and Ross pushing back hard on some points here.
podcasts.apple.com/us/podcas…
Every time I start writing a new essay, I worry that I’ve read so much AI-generated content that AI-speak is silently infecting my subconscious and seeping into my writing.
For the past couple of months, I've been quietly building at Vetto what I believe is essential for the next generation of model evaluation and post training.
If the pace of model development is relentless, so must be the data and evaluation infrastructures that fuel it.
As benchmarks demand greater headroom and fairness, keeping pace through traditional services is increasingly unsustainable.
Computer Anthology is the premiere of our proprietary, scalable engine towards automating an evaluation system that consistently differentiates the frontier.
gettrack.link/r/GoUEbYDC
We recorded our full LP presentation on Decade, with the best clips from Vitor's interview included. Happy to DM it privately, just ask for it in the comments.
Congratulations to Vitor, Felipe, and the whole Decade team for your public debut. We are excited for this adventure!
The job of a venture investor.
We spend years building frameworks that guide how we invest, but earning our keep means knowing when to throw them out the window.
We had not gone pre-product in six years and nearly 30 investments. We have now backed Latin America's most valuable AI companies. When the team is this good, sitting it out is the real risk.
Vitor has been here before. Surfing the biggest tech waves of the moment. Building a company 100% around the technology.
Nubank went all in on mobile and the cloud back when incumbents saw them as features. Decade is doing the same with AI, building the entire company around it from the first line of code.