Facundo Franco

@rauchg whatsapp-sales.vercel.app WhatsApp sales agent for e-commerce. Converts high-intent conversations into purchases automatically. Built with Claude API + Node.js/Express.

@ahsan_fayaz1 @rauchg Passes cost directly to the client as part of the monthly fee. Margins stay clean.

@facundofranco_ @rauchg How much are you paying for claude api?

Show me the thing you’ve built with AI you’re most proud of. Reply with a working product URL and what model / agent you primarily used.

@RoundtableSpace the operator layer. Systems that make AI agents actually work in production past week one.

What are you building today?

Conversations are where e-commerce sales get lost. Sella closes them. Built with Claude API. Running in production.

@rauchg whatsapp-sales.vercel.app WhatsApp sales agent for e-commerce. Converts high-intent conversations into purchases automatically. Built with Claude API + Node.js/Express.

@inetimiomu @rauchg More than a bot. It understands the conversation, recommends the right product, and pushes to checkout. Built on real e-commerce conversations.

@facundofranco_ @rauchg Interesting, so it’s a WhatsApp sales bot?

@aakashgupta The alpha moved from processing information to curating taste. That's the whole shift in one line.

The CPO of a $131M AI company just confirmed that PMs at AI native companies are now indistinguishable from engineers. Then she showed exactly why. She built a PM agent from one prompt in Claude Code. It pulls GitHub issues, scores every single one by priority, generates a daily report of what to build next. Instrumented the entire thing with one command. No IDE opened. No engineering partner. Traces streaming into her observability platform within minutes. Then she showed the self-improvement loop. The agent evaluates its own scoring accuracy, identifies categories where it misjudged priority, and feeds corrections back into itself. Bugs were getting scored too low. The agent caught the pattern, flagged it, suggested the fix. That cycle runs on a cron while you sleep. Her analogy was tennis. Nadal studies his own plays to get 1% better every day. Self-improving agents study their own traces. The PM's job used to be consuming more user feedback than anyone else. The agent now consumes all of it. Every GitHub issue, every Gong call, every Slack thread. What's left for the PM is the eval. Defining what "good" looks like. Deciding that bugs always outrank new features. Deciding which customer pain matters most. The alpha moved from processing information to curating taste. She confirmed same-day shipping is already happening. Issue comes in, PM identifies it, Claude Code prototypes the fix, ships that afternoon. The PM who manually scans a prioritized backlog every morning is competing against a PM whose taste agent runs 24/7 and improves itself overnight. Any PM running observability and evals on their agents is probably already in the top 1%. Given what this workflow produces, that tracks.

@RoundtableSpace The bottleneck was never the agent. It was always the coordination layer.

THE CREATOR OF CLAUDE CODE SAYS SINGLE AGENTS ARE ALREADY OBSOLETE • Boris Cherny explained why the future is coordinated teams of AI agents, not better prompts • One agent researches, one builds, one reviews, while another orchestrates the workflow

@RoundtableSpace You can outsource your thinking but you can't outsource your understanding. That's the whole talk in one line.

KARPATHY JUST EXPLAINED THE FUTURE OF SOFTWARE FOR FREE • Anthropic reportedly hired Andrej Karpathy while he simultaneously dropped a free 29-minute AI talk • Covers Software 3.0, why “vibe coding” is evolving, and how LLMs actually behave

Modal just raised $355M at $4.65B. The compute layer is getting funded at scale. The next question is who builds the workflows that run on top of it. That’s the operator.

@RoundtableSpace 382K downloads on day one. Small businesses don't need to hire, they need to install.

ANTHROPIC'S 31 SMALL BUSINESS SKILLS GOT 382,000 DOWNLOADS ON DAY ONE AND SOMEONE JUST MAPPED EVERY SINGLE ONE INTO A 10 MINUTE SETUP. It covers financial operations, sales, HR, marketing, and reporting with a full connector guide and real output examples.

@gdb Best time to be building. The floor keeps rising.

GPT-5.5 is a very good model

@tbpn @danshipper AI makes the work cheaper to start. It doesn't make the judgment cheaper to finish.

At Every, all employees use agents + have unlimited token budgets. Since GPT-3, they've increased headcount from 4 to 30. "If you're actually on the frontier like we are," @danshipper said, "it seems like there's more human work to do than ever."

@garrytan The manual phase isn't inefficiency. It's data collection.

Everyone thinks "do things that don't scale" is about building relationships with early users. Yes AND it's about generating mistakes at maximum density. When you're doing everything manually (onboarding, support, delivery) you hit errors every hour. Each error teaches you something the dashboard never will. The manual work IS the learning. Automate too early and you freeze your ignorance in code (and now markdown).

@garrytan Most people are still treating the harness as an afterthought. It’s the actual product.

My simple secret to agentic coding forbes.com/sites/josipama…

@levie Every bottleneck AI removes reveals the next one. The review layer keeps winning.

Here’s a key line in this mythos update. This is precisely an example of why engineers don’t go away, ever. We’ve made it far easier to create and find security issues, which means the new bottleneck is our ability to actually review, respond to, and fix the issues. Far from AI magically solving all of this, there still is major triage work and human judgment required to do the follow on work to actually protect systems. As a result, we’re about to enter a security engineer boom. Jevons paradox all over again.

I dropped a job board link into Claude. First response was linear: here are the two roles that fit you. I pushed back. Asked what the data actually meant. That unlocked the real insight: 186 internal AI roles posted in one day, no agreed title, category forming in real time. That became the post. Not the jobs. The pattern behind the jobs. The skill isn't using AI. It's knowing when to push past the first answer.

@tjack The deployment gap is bigger than the model gap right now.

Great FDEs are worth their weight in gold right now. Just saw my first competitive acquihire (multiple term sheets) to land a very talented team - not to lead Eng, Product or Design...but FDE. Shows you where one of the biggest talent bottlenecks is.

186 internal AI roles in one day. Not customer-facing products. Internal. Companies hiring people to build AI for themselves first. Nobody agrees on what to call the role yet. That's what a category looks like before it has a name.