Brief (a16z sr005)

AI coding tools ship with no way to give agents product context. So agents get kept in the dark. Fed whatever's in the repo. Expected to produce anyway. They excel at migrations and refactoring. They struggle with feature development. Not because they're bad at coding, but because feature work requires context that isn't in your codebase: Who are your users? What does success look like? What have you already tried and rejected? Brief gives your agents this context. Decisions, personas, constraints. So they stop suggesting things that miss the point.

That code you rejected 3 times? Your AI just suggested it again. It's not a capability problem. It's a memory problem. AI coding agents start every session fresh. The corrections you made yesterday? Gone. The design system you explained? Forgotten. The architectural decisions from last month? Never knew them. This is why we built Brief. Your decisions, constraints, and context persist across sessions. The agent remembers what you've already told it.

Avi Chawla just published a solid breakdown of the .claude/ folder. claude markdown, custom commands, skills, agents, permissions. claude .md is the most important file in your repo that nobody owns. Teams are spending real hours engineering their .claude/ folders. That's genuinely valuable. But claude .md is a config file. It tells your coding agent how to behave. It doesn't capture why you made the decisions you did, which customers drove those tradeoffs, or what you decided NOT to build. The moment you have more than one engineer (or more than one AI agent), that institutional memory lives in someone's head. Or nowhere. We're building Brief to solve this. Brief connects to Claude Code (and Cursor, Windsurf) via MCP. When your agent is about to build something, it calls ask_brief to check: does this align with what we've already decided? Does this serve the right customer segment? Did we already ship something that does this? claude .md tells your agent how to write code. Brief tells your agent whether it should be building this at all. If you're investing in your .claude/ setup, Brief is the layer above it.

MCP gave AI models a way to call tools. But it assumed one model doing all the thinking. When you need multiple agents working together, that breaks down. The orchestration logic, error handling, context passing, you build it all from scratch. Every team reinvents it differently. ACP (Agent Communication Protocol) treats agents as first-class citizens in a network. REST-based, async-first, with built-in discovery. Agents register what they do, orchestrators look them up dynamically, and long-running parallel workflows actually work instead of timing out. The hardest part isn't the protocol though. It's context. Agents don't share memory. Each one starts cold. The quality of your context-passing architecture determines whether your agent network behaves like a coherent system or isolated components talking past each other. Wrote up the full breakdown of what ACP does differently and why it matters for production agent systems. Link in comments.

 "Show me WAU since January" That's what Stephan Castro typed. Brief pulled the data from PostHog and rendered an interactive chart, right in the conversation. No switching to a dashboard. No writing SQL. No exporting to a spreadsheet to make it legible. Just a question and an answer you can actually see. We just shipped inline data visualizations inside Brief. Ask a question about your product, get a chart back. Line graphs, bar charts, funnels; rendered live, with the context your AI already has. Brief is AI product infrastructure for early-stage teams. And good infrastructure doesn't make you open four tabs to answer a simple question. It just shows you the chart. 🔗 in the comments

Brief just shipped a lot. Here's the rundown. ⚡  🤖 Brief Slackbot The Slackbot has evolved into something genuinely powerful: a multiplayer AI experience where Brief acts as strategy participant, decision recorder, and orchestrator. Right inside the conversations where work already happens.  📊 Market Intelligence Brief now understands your market: - Revenue & Pipeline. Connected to Stripe, HubSpot, Attio, and Lightfield, so Brief knows the dollar value behind every feature request.  - Competitive Intelligence. Brief tracks competitors, builds battle cards, and watches for changes in products, pricing, and messaging.  - Positioning. Messaging pillars, differentiators, and proof points, all in context.  🖼️ Wireframes & Flowcharts Generate wireframes and flows inside Brief chat and docs. This week, Claude Code one-shotted a complex user flow. It had the right visual context to work from. More updates:  → Brief CLI (ask_brief). AI agents can pull product context and get feedback from their own 24/7 robo PM.  → Sub Agents. Brief parallelizes work and maintains context across a fleet of agents.  → Teachable Skills. Save any workflow as a reusable skill. Modify Brief's personality via its identity file.  → Database Integration. Supabase now, Postgres next. Brief reads usage data to make smarter product decisions.

14% of workers using AI report burnout. 39% more errors. 34% intend to quit. The cause: reviewing AI output at the wrong altitude. You open a 200-line diff. Scan every line. Try to reconstruct the intent behind each decision. Wonder if the agent understood the billing API constraint. Trace a suspicious pattern back through three files. You're reverse-engineering decisions from syntax. That's exhausting. The fix: give your agents context before they write code. Review at the decision level, not the syntax level. full breakdown in the comments

Your team has made thousands of product decisions. "We're not building that feature." "Use the existing design system colors." "That exception only applies to enterprise." They live in Slack threads. Meeting notes. Comments buried in Linear tickets. Your AI coding agents (Cursor, Claude Code, Windsurf) have no idea they exist. So they build things you already decided against. Suggest patterns you've explicitly rejected. Miss context that would make them useful instead of just fast. Some teams solve this with infrastructure: Vector databases. RAG pipelines. Versioned design tokens. CI enforcement layers. System prompts injecting org rules dynamically. It works. It also takes months to build and a full-time job to maintain. Instead, just run it by Brief. Brief captures your product decisions from Slack, Linear, Jira, and Notion. All automatically, without changing how you work. Then it pipes that context directly into your AI agents via MCP or CLI. You make a decision. Brief remembers it. Your agents respect it. Setup takes minutes.

The first product hire at a startup has the hardest onboarding in tech. They have to inherit a founder's brain. Six months of "we almost built X but didn't because of Y." Three pivots that happened in Slack threads. A pricing decision made on a call that nobody wrote down. Customer patterns that only the founder can pattern-match in real time. The first 90 days? Archaeology. This is the type of problem Brief was built for. When a founder uses Brief, they're building institutional memory as they work. Every decision captured. Every tradeoff explained. Every "why we built it this way" stored and searchable. For teams moving from "founder does product" to "founder + PM does product," this is the transition that usually breaks things. Brief makes it survivable.

Shipping a feature used to take weeks. Now it takes an afternoon. AI coding tools and no-code platforms gave every founder the same superpower. Speed stopped being the advantage. Time to product-market fit became the constraint. "Build a bunch of things until something sticks" just got commoditized. Every team is trying that strategy. The result: thousands of fast, forgettable products that die after launch. The founders who reach PMF aren't the fastest shippers. They understand their users deeply enough to know what makes someone buy, come back, and refer. Every touchpoint has to earn its place. The onboarding flow. The support interaction. The email. The moment someone realizes they need you, not just the feature itself. Speed is table stakes. Customer insight is the new moat. Your best customers bought for a reason. Find out why they chose you over doing nothing, and you'll reach PMF faster than the teams still spamming features.