Pierre Schurmann (@Xurmann) profile picture

CEO Nuvini. Building AI into real LatAm B2B software. Measuring ROT — Return on Tokens. Brazil → world

Brazil
Joined October 2009
Photo by Pierre Schurmann (@Xurmann)Photo by Pierre Schurmann (@Xurmann)Photo by Pierre Schurmann (@Xurmann)Photo by Pierre Schurmann (@Xurmann)Photo by Pierre Schurmann (@Xurmann)Photo by Pierre Schurmann (@Xurmann)Photo by Pierre Schurmann (@Xurmann)Photo by Pierre Schurmann (@Xurmann)Photo by Pierre Schurmann (@Xurmann)Photo by Pierre Schurmann (@Xurmann)
Pierre Schurmann retweeted
We’re hiring a Growth Lead for Brazil, based in São Paulo. 🇧🇷 This person will help shape how people across Brazil discover ChatGPT, find value in it, and keep coming back. The work spans product growth, experimentation, and performance marketing, turning local user insights into better product experiences and measurable growth. If you’ve built and scaled growth programs, understand Brazil’s users and culture, and enjoy getting hands-on with both product and marketing, take a look. Apply here, or share with someone who would be a great fit: openai.com/careers/growth-le…
45
40
10
522
30,351
Pierre Schurmann retweeted
Dois brasileiros fazendo sucesso globalmente. @resend @firecrawl 🇧🇷
how do you turn a repo with 1,000 stars into a product used by 1.2 million developers? @firecrawl co-founder and CTO @nickscamara_ walked me through it. two Brazilians in SF, one camera, and a lot to say about agents.
1
12
430
Pierre Schurmann retweeted
We've added `ant apply` to the ant CLI. Now you can declare Claude Managed Agent environments, agents, skills, memory stores, and deployments as files in your repository and keep the API's resources in sync with them using `ant apply`. Docs: platform.claude.com/docs/en/…
153
163
33
2,564
371,332
Pierre Schurmann retweeted
prev startup advice: do things that don't scale 2026 update: do things that don't scale, then build AI agents to scale them
55
25
5
297
12,485
Pierre Schurmann retweeted
It’s pretty clear that the emerging paradigm of agents will be like if you had a human expert in any domain, and they had all the capabilities of a top engineer who could use any tool (or the write their own on the fly) to complete any task, along with unlimited compute and a file system to work with. That combination of skills and technology primitives provides you with somewhat limitless capability in AI. You’re no longer limited by only what the model was trained on, or the inherent context window limitations. The agent will simply spin up subagents to work on component parts of the workflow, and get expertise as needed throughout the process. For all known types of tasks that are frequently repeated, they have quick access to existing skills and tools to complete their work. We’re already seeing this in a range of fields where skills are being written for agents to follow either domain-wide or company-specific processes. Doing legal analysis in a specific way, running financial models, processing spreadsheets for complex data work, generating PowerPoints, and so on. And for areas they’ve never seen before, they can simply write code on the fly to do the work one-off. Imagine pairing an industry expert with an engineer that can code up any custom script whenever it wants. Compute is your only limiter. This approach seems to cover a fairly wide range of knowledge work. Obviously the first space to benefit the most from this has been in coding itself, but it’s clear that this go across all other areas of work and even personal agents. Kind of wild.
78
64
16
535
101,767
Pierre Schurmann retweeted
Comfort is seductive because it feels harmless, but it quietly compounds into weakness
68
207
18
1,891
99,886
Pierre Schurmann retweeted
We just shipped 32 new integrations — our biggest update yet. Outlook, Teams, SharePoint, OneDrive, Google Chat, BigQuery, Stripe, Canva, Ramp, Attio, Mixpanel, and 21 more. If your team uses it, Tasklet can connect to it. 🧵
4
6
6
27
9,247
Pierre Schurmann retweeted
Replying to @LLMJunky @karpathy
LLM Council Skill: github.com/am-will/codex-ski… Install to Codex: npx @am-will/codexskills --user am-will/codex-skills/skills/llm-council
4
6
61
5,153
Pierre Schurmann retweeted
Running a company is just context engineering internally. Now that skill has even more value in the agentic world. Us tech founders have been doing reps to prepare for this.
135
161
36
2,581
150,744
Pierre Schurmann retweeted
Best articulation I’ve seen on why high agency is less common than you’d expect. from @george__mack
30
326
29
2,902
120,266
Pierre Schurmann retweeted
We've spent a year helping companies figure out AI. The biggest thing we've learned: most teams aren't stuck because they lack tools. They're stuck because they don't know where to start. So, we built a framework with four levels of AI adoption—from using ChatGPT for tasks to running agents that handle entire workflows. Today we're sharing everything: • What separates Level 1 from Level 4 • The tasks that automate well • A simple way to get started (write a job description for your AI intern) And we're opening up a limited number of consulting spots for Q1. Full breakdown from @NataliaZarina: every.to/on-every/the-next-c…
2
6
1
97
12,414
A PROPOSAL FOR UNIVERSAL HIGH INCOME (UHI): During my recent Moonshots podcast with @elonmusk, we dove into his notion of Universal High Income (UHI) – Elon’s proposal that an AI and Robotics will enable a world of sustainable abundance for all... a life beyond basic income, towards high income and standards of living. When I asked him how this might work, he said: “You know, this is my intuition but I don’t know how to do it. I welcome ideas.” That single statement has been ringing in my head ever since. Here’s why: the economics of scarcity are flipping to the economics of Abundance. I do believe that AI and humanoid robots can produce nearly anything we need—goods, services, healthcare, education—at costs approaching zero. But there’s a gap between that vision and getting there. How do we actually fund and distribute Abundance to everyone? Today, I’m excited to share one compelling answer. I’ve been talking to Daniel Schreiber, CEO of Lemonade (the AI-insurance company that just launched 50% off premiums for Tesla FSD drivers), about a framework called the MOSAIC Model: a concrete proposal for how governments could implement Universal High Income without raising taxes on workers or businesses. (See the components of MOSAIC in my P.S. below.) Here’s the core insight that makes the math work: 1/ THE AUTOMATION PARADOX: AI Unemployment ≠ Traditional Unemployment When most people hear “mass job displacement,” they picture economic collapse: bread lines, depression, social chaos. That’s because they’re thinking about traditional unemployment, where workers disappear and nothing replaces them. AI unemployment is fundamentally different. Think of it this way: imagine sending a digital twin to work in your place. It performs your tasks faster, cheaper, and better. The company’s output increases. GDP grows. The resources exist – they just need to be redistributed. This is the Automation Paradox: AI can raise productivity while displacing labor. When workers are replaced by more productive capital, GDP rises even as fewer humans work. The challenge is not affordability. It’s capture and distribution. 2/ “AI DIVIDEND”: Where the Money Actually Comes From Daniel’s framework identifies two places the AI surplus shows up, and how to capture it without disrupting consumers or raising statutory tax rates: Channel 1: Dynamic VAT (The Deflation Dividend) AI is deflationary. When AI cuts the cost of producing something by 30%, that value creation can either flow entirely to shareholders – or be partially recaptured for society. Dynamic VAT works like this: as AI drives quality-adjusted price declines in goods and services, the VAT rate adjusts upward by exactly enough to keep consumer prices stable. Consumers pay the same. But the government captures part of the deflation dividend. It’s frictionless redistribution. Prices don’t rise. No one feels it. Channel 2: Over-Trend Profit Ring-Fencing AI is generating windfall profits for companies at the frontier. Rather than raising corporate tax rates (which drives capital flight), the MOSAIC Model proposes ring-fencing only the above-trend portion of capital income tax receipts. Baseline profits? Untouched. Normal corporate taxes? Unchanged. But what about the incremental surge in profits attributable to AI? A portion gets earmarked for the “Universal High Income” fund. Statutory rates stay the same. Companies keep most of their windfall. But society captures enough to fund a universal floor. 3/ WHAT THIS MEANS FOR FAMILIES: Here’s where it gets real. Under the MOSAIC Model’s basic implementation (before any additional policy choices), a household with two non-working parents and two children would receive income equivalent to today’s fourth decile: roughly the 30-40th percentile of current household income. To be clear, that’s not survival-level subsistence. It’s lower-middle-class security. For doing nothing. This creates a Universal Basic Floor – funded entirely by the two low-friction channels above. But this is just the starting line, not the finish line. If society chooses to capture more of the AI dividend through additional mechanisms (windfall levies, land-value capture, AI-services taxation), the floor could rise to what Daniel calls the “the UHI Benchmark”: approximately 120% of median wages. Upper-middle-class income. Universal. The surplus exists. The question is: how much do we collectively choose to redistribute? 4/ WHY TIMING IS EVERYTHING: Here’s what keeps both Daniel and me up at night: the political window for implementing this is closing. The MOSAIC Model’s political economy analysis shows something counterintuitive: feasibility is highest early in the AI transition – before capital consolidates opposition, before tech incumbents organize billion-dollar lobbying efforts, before the status quo hardens. Wait until mass displacement is undeniable? By then, it may be too late to pass anything. Act early or not at all. A good system passed in 2026 beats a perfect system proposed in 2030 that fails. 5/ THE INVITATION: Elon said he welcomes ideas. This is one. The MOSAIC Model isn’t the only answer, but it’s a rigorous, economically grounded starting point. It demonstrates that Universal High Income is not utopian dreaming. It’s an engineering problem with identifiable solutions. The AI dividend is real. The fiscal math works. The question is whether we have the collective will to build the capture mechanisms before the window closes. The full MOSAIC Model is available today at mosaic.org.il/model for policymakers, economists, and fellow entrepreneurs to critique, improve, and implement. Read the full plan, verify the math, and let’s debate this. Because this is not a matter of any single country or company getting it right. It’s about humanity navigating the biggest economic transition in history. When AI takes our jobs, it should also pay our wages. Let’s make that happen. Peter Diamandis (in collaboration with Daniel Schreiber, @daschreiber, CEO of Lemonade and Chair of the MOSAIC AI Policy Institute) P.S. The detailed components of MOSAIC that make the model affordable: M – Multi-channel / Mechanism (Implied): The core philosophy that no single tax can fund UHI alone; it requires a “mosaic” of multiple bases. O – Over-trend Ring-fencing: Earmarking 85% of the “windfall” capital-income tax receipts (profits and capital gains) that exceed historical trends. S – Savings (Government Automation Dividend - GAD): capturing the cost savings from automating government bureaucracy (e.g., using AI for back-office admin). A – AI-linked Deflation (Captured via Dynamic VAT): The largest tile. As AI drives prices down, the VAT rate adjusts upward to capture the “deflation gap,” keeping prices stable for consumers while generating revenue. I – Income (Negative Income Tax): The distribution mechanism itself, ensuring work always pays. C – Consolidation: Rolling existing, overlapping welfare transfers into the new single payment to avoid double-spending. In short: The MOSAIC is the Fiscal Architecture. It argues that while one tax (like a “wealth tax”) is politically impossible or insufficient, a mosaic of VAT + Windfall Profits + Efficiency Savings + Legacy Consolidation creates a robust funding base for a poverty-ending income floor.
419
277
94
1,760
320,449
Pierre Schurmann retweeted
"Value your time. It is all you have. It’s more important than your money. It’s more important than your friends. It is more important than anything. Your time is all you have. Do not waste your time." @naval
97
511
18
3,196
187,983
Pierre Schurmann retweeted
If you push it you will see all the ways it breaks but there is a powerful fix: put it in Claude.md and it actually helps quite a lot Simple things like tell me if it’s overengineered, underengineered or just right when exiting plan mode. Write the failing test on a production bug before you fix it.
Replying to @karpathy
As always, a very thoughtful and well reasoned take. I read till the end. I think the Claude Code team itself might be an indicator of where things are headed. We have directional answers for some (not all) of the prompts: 1. We hire mostly generalists. We have a mix of senior engineers and less senior since not all of the things people learned in the past translate to coding with LLMs. As you said, the model can fill in the details. 10x engineers definitely exist, and they often span across multiple areas — product and design, product and business, product and infra (@jarredsumner is a great example of the latter. Yes, he’s blushing). 2. Pretty much 100% of our code is written by Claude Code + Opus 4.5. For me personally it has been 100% for two+ months now, I don’t even make small edits by hand. I shipped 22 PRs yesterday and 27 the day before, each one 100% written by Claude. Some were written from a CLI, some from the iOS app; others on the team code largely with the Claude Code app Slack or with the Desktop app. I think most of the industry will see similar stats in the coming months — it will take more time for some vs others. We will then start seeing similar stats for non-coding computer work also. 3. The code quality problems you listed are real: the model over-complicates things, it leaves dead code around, it doesn’t like to refactor when it should. These will continue improve as the model improves, and our code quality bar will go up even more as a result. My bet is that there will be no slopcopolypse because the model will become better at writing less sloppy code and at fixing existing code issues; I think 4.5 is already quite good at these and it will continue to get better. In the meantime, what helps is also having the model code review its code using a fresh context window; at Anthropic we use claude -p for this on every PR and it catches and fixes many issues. Overall your ideas very much resonate. Thanks again for sharing. ✌️
24
15
2
283
86,040
Pierre Schurmann retweeted
So much doom and gloom in the world but then you meet young founders and it’s just impossible to not be optimistic about the future.
6
7
135
21,074
Pierre Schurmann retweeted
"It's lovely to be useful." — Jony Ive
97
250
26
2,988
237,428
Pierre Schurmann retweeted
VC firm 500 Global will invest $300K in each of 30 AI startups in Latin America—a major jump from its previous $80K. Founders will get 12 months of mentorship from 700+ experts and hands-on support in product, sales, and marketing...
1
1
1
3
622
Pierre Schurmann retweeted
Anyone know why ChatGPT still can't see your screen? I would love to have voice mode going and just have it coach me through a technical problem. (Yes, I know it can watch your Terminal, Cursor, etc)
36
1
2
92
24,296
Pierre Schurmann retweeted
Your new self will cost you your old self.
179
147
13
1,197
49,825