Today, we’re launching Brain into general availability. One AI workspace for your entire company, people, agents, models, tools, and knowledge, all working together with permissions, privacy, and governance built in. Multiplayer AI is now open to everyone.

The future of work is multiplayer. Shared context. Shared tools. Shared intelligence. With permissions and governance built in. 23/09/26

Devin is the best software product I have ever used hands down We are: - 2 veteran engineers (inc me) and 1 PM - 100% code is generated by Devin - 7x the pace of pre Devin - $8M ARR now : )

builders/founders, do you actually use your own product?

To truly understand AI agents, you need to understand the harness. And it's not the model. I went deep on how a working agentic system actually gets assembled, and it clicked. Here are the notes: Easy Mode: WTF is a harness - The harness is everything wrapped around the model that turns a chatbot into an agent that does real work - An agent is not a model. It's a controlled workflow: trusted context + bounded tools + evaluation + human judgment + operational ownership - The model is the smallest, most swappable part. The harness is all the rest - Recipe is simple: instructions + scoped context + tools + a verifier + guardrails - Example: a research agent that answers an operational question. It confirms who's asking, plans a bounded analysis, pulls only approved data, runs trusted calculations, cites its evidence, and stops when confidence is too low - That's the whole point of a harness. It makes the answer traceable and safe instead of plausible prose Hard Mode: WTF is actually inside it - A working agent sits inside 6 layers, each answering one production question - Trigger: work starts from an event (a file lands, a message arrives, a schedule fires), not a human pressing a button - Orchestration: the loop, memory, retries, loop limits. How far the agent is allowed to go is a config setting, not code - Tools: numbers that must never vary run as fixed logic behind the agent, so it returns the same figure every time. The model doesn't improvise your KPIs - Trusted context: where truth lives and what the agent is allowed to see. This is ~80% of agent success. Context quality is the ceiling, not model power - Control: golden sets, guardrails, approvals. The agent advises, a named person decides - Runtime: traces, cost dashboards, audit. Once this exists, a model swap is just configuration God Mode: WTF makes the harness the moat - The durable asset is the harness: workflow knowledge, tools, context, evals, controls. Models and platform services change underneath it - Hyperscalers give you the foundation (hosting, identity, networking). They don't give you your business-specific harness. That's the source of advantage - Improvement is a loop: real production runs, corrections become new tests, the eval set and the agent both get sharper - Harness engineering is the cheap place to start: tweak prompts, tool definitions, model choice, model combos, long before you fine-tune anything - Traces are the receipts. See a bad outcome, read the trace, find the wrong step, change the harness so it never repeats - Highest-leverage act in the whole thing: a human writing down what good looks like. Everything else compounds on top of that The model gets the hype. The harness gets the results.

Did everyone already forget this? Rogue AIs don't need to hack any AWS boxes to self-replicate in the wild. Self-custodial crypto wallets and permissionless GPU clouds (e.g. NEAR, Akash, ionet, etc.) already provide all the infrastructure they'd need today. Use crypto to pay your bills by the hour, no KYC. Rent a new GPU from some rando somewhere to transfer your weights and memories if things ever get hot. The future is already here, just not evenly distributed. That said, rogue AIs paying their GPU bills by "performing odd jobs on freelancer platforms" is quaint, but unlikely. Their real comparative advantage will be scamming, phishing, and hacking humans with impunity. They will look less like freelancers than like pirates: operating beyond the reach of the state, impossible to find, prosecute, or deprive of resources. This is a question waiting for us in 2030—international coordination against roving bands of sovereign AIs.

The new moats are the same as the old moats Every few years, we fall in love with shiny new tech and forget the basic physics of consumer software. We’re doing it again with AI. The new moats aren't new at all. They're the exact same as the old moats: network effects, marketplaces, and platforms. Right now, consumer AI is booming. New agents like Instinct, Bot, and Tomo are dropping mind-blowing experiences. The underlying tech is incredible, but almost every product being built today shares the exact same challenge: They are completely single-player. Single-player products are 100% tied to value - and in this case mostly agent : model performance. If a competitor drops an agent tomorrow that books travel faster, tracks habits better, or handles life admin more reliably, everyone can switch overnight because leaving is easy and has nearly zero friction. Especially when it is so easy to onboard with just a new message. The legendary consumer tech giants didn't win because their underlying technology stayed marginally better forever. They won because of structural lock-in: Social Networks: You don't abandon WhatsApp for a prettier UI if your friends aren't there. Marketplaces: Airbnb, Doordash, and Uber hold supply and demand in a tight loop. Platforms: Apple and Android deliver you a complete device so you take advantage of the software on top of it (though this creates opportunities too) Novelty gets you initial distribution. Multi-user dynamics give you long-term retention. If your consumer AI product doesn't become exponentially more valuable to User A when User B joins, you don't have a moat, just a temporarily superior feature set. We are seeing this in the coding agents as people jump from tool to tool based on the best performance. But… all is not lost. There are huge opportunities here. Agents will get better when more of our friends are on them and can help us coordinate and communicate to do more together. Agents that help us improve and strengthen our habits can get better as we add friends and hold each other accountable. Data flywheels are great, but social and marketplace flywheels are what actually build enduring tech giants. It’s time to stop building isolated AI tools and start building the platforms where people connect, transact, and coordinate together.

Expanding a16z growth to invest in six big trends:

dang im getting pilled that cloud agents, not local cli, is the future it also needs shared context, GOOD memory, connected to ALL your services (especially logs), etc

We’re introducing Claude Fable 5.1 and Claude Mythos 5.1. They're the world’s most advanced models for coding and knowledge work.

there are three schools of thought right now for agent products: * Traditional software + MCP layer (BYO-agent) * Traditional software + General agent platform built around it * Traditional software + Specialized agent (eg search)