Federico J. Botti retweeted
Workflow SDK 5 and Vercel Workflows are a monumental step forward for agents.
◾ The 'time to first step' and 'time between steps' have massively improved.
◾ We doubled down on encryption and security.
◾ The reliability of the overall system has been proven at massive scale. We've ironed out many subtle details and exercised many disaster-recovery scenarios.
◾ We made the underlying durability log reusable for any agent harness. 𝚍𝚞𝚛𝚊𝚋𝚕𝚎-𝚑𝚊𝚛𝚗𝚎𝚜𝚜 was born.
Check it out:
Workflow SDK 5 is built for fast agent loops.
✓ Up to 25x lower latency and overhead
✓ Every payload is encrypted by default
✓ Regional workload isolation & failover
vercel.com/blog/workflow-sdk…
New in agent-desktop - sandboxes 📦
Your agents can now install, create, start, and drive their own macOS VM, or run in a remote Sandbox
if you're on a mac and want computer use without the agent touching your personal stuff, here's how i do it
run agent-desktop inside a macOS VM. Apple's Virtualization framework boots a full guest in a window, and the agent only sees what's inside it.
you keep working on your mac and your agents runs inside the VM
Federico J. Botti retweeted
New experiment: Your agent can now save a browser task as a flow and run it again later with new inputs, without a browser
→ "Check today's price on this product"
→ "Look up an order's status"
→ "Pull this week's numbers from the dashboard"
Almost no tokens, instant
What does one AI conversation actually cost?
Most founders scaling an AI product can't answer that.
I wrote up the real numbers: tokens, caching, routing, and the hidden costs nobody budgets for 🧵
The viability check is simple:
cost per conversation × conversations per user per month
must sit well below what the user pays.
Track cost per completed task, not just per API call.
1
3
The real problem isn't that AI is expensive. It's not knowing what it costs.
Full breakdown, plus the 8 levers we use at Persiscal:
persiscal.com/en/blog/real-e…
1
4
Wow! Moving from agents that use tools to agents that can operate an entire computer—in parallel, without interrupting you. Just imagine the workflows this unlocks.
This also can be used with agent-browser.dev/
Introducing agent-desktop
agent-browser, for your whole computer
→ Computer Use for any coding agent
→ macOS, Windows and Linux
→ Runs in bg while you keep working
→ Agents use your computer in parallel, each with its own cursor
25
Federico J. Botti retweeted
Introducing agent-desktop
agent-browser, for your whole computer
→ Computer Use for any coding agent
→ macOS, Windows and Linux
→ Runs in bg while you keep working
→ Agents use your computer in parallel, each with its own cursor
Federico J. Botti retweeted
Star it, fork it, tell your agents about it
Built by @lahfir, who just joined the Vercel Labs team!
github.com/vercel-labs/agent…
Federico J. Botti retweeted
We’ve raised $1.3M from YC & Afore Capital so you never have to type or speak again.
Typing is slow. Talking into your computer is awkward. Subvocal fixes both
Our glasses give you the speed of speech without you making a sound or touching a keyboard.
It tracks your internal speech muscles with non-contact sensing and decodes silent, almost invisible mouthing into text at more than 200 words/min.
Pre-orders are now live at:
subvocal.co/preorder?utm_sou…
Federico J. Botti retweeted
It's incredible to watch a new kind of economy be born. Agents purchasing infrastructure products and services. We've been shocked by how often they buy from our marketplace via our CLI already. We're now extending this to domains.
Agents can now go full stack, from idea to online business, with a banger domain name 😁
Federico J. Botti retweeted
Muse, Instinct, OpenClaw, Claude Code…
All successful agents have 3 key components:
🧠 Brain → model, harness (logic)
👐 Hands → tools, computer, browser
🗃️ Files → memories, skills, repos
The 'easy' way is to throw all these in 1 stateful computer (a Mac Mini)
Like, you run 𝚌𝚕𝚊𝚞𝚍𝚎 or 𝚏𝚡 in your mac, you keep it running all day with 𝚌𝚊𝚏𝚏𝚎𝚒𝚗𝚊𝚝𝚎, it has storage, and CLIs and apps installed.
But if you want to cost-efficiently run agents in the cloud, you actually start breaking down these parts.
🧠 The harness can run in Fluid compute. To make it reliable across restarts, rollouts, crashes, you make its event log durable using Workflow.
👐 The hands can be a dedicated browser fleet like Browserbase/Kernel, a computer like Sandbox, and even more efficient lightweight tools like just-bash.
🗃️ 🆕 What was missing was a way to also decouple storage. Imagine you want to run a memory consolidation cron job every night ("dreaming"). You can read/write to the files directly without 'booting up' the agent's full computer.
Today we're introducing the perfect companion to Sandbox: Drives. We shipped the computer for agents, now we're giving you the 'external disk' you can attach at will. It's early, and we'll be expanding capabilities here quickly.
Btw, breaking apart the agent into these independent parts not only optimizes costs in a big way, it also *massively* improves security and auditability. I'd argue you can't even run a secure agent otherwise!
Vercel Sandbox now has persistent storage with Drives, in public beta on every plan.
▪︎ Store agent workspaces, data, models, deps
▪︎ Read snapshots across parallel sandboxes
▪︎ Mount up to four Drives per sandbox
▪︎ Up to 16 TiB per Drive
vercel.com/changelog/drives-…
Federico J. Botti retweeted
358 sign ups en el primer GrokBot Meetup Buenos Aires que organizamos con @uTombou el miercoles pasado.
Lamentablemente tuvimos que dejar mucha gente afuera por temas de capacidad, pero ya vamos a estar organizando mas eventos
Algunos highlights de la tarde:
- Abrio el evento Ria (Field Engineer @spacexai)
- Siguio Robert Di Marco (Engineer @spacexai) donde mostro como usa GrokBot y tambien dio un espacio de Q+A donde le preguntamos de todo
- Las oficinas de @SantexGroup estuvieron INCREIBLES.
Para los que quieran ver el evento, quedo grabado aca youtube.com/live/SBn6doEYZbQ…
Los esperamos la proxima!
Federico J. Botti retweeted
self-driving en Argentina HOY 🇦🇷
ba -> mardel 89% de los kms autónomos con el comma 4 en un toyota corolla NORMAL
387 km o 3:50h manejados por openpilot full autonomo
31 kms o 41m totalmente manual por humano
8 km o 5 mins acelerando humano y volante open pilot















