Been playing with Claude Code mods.
3 I'm running right now:
- restate-intent: when I dictate a long prompt, Claude first restates in a sentence or two what it understood my goal to be, then carries on without waiting. It only fires on long dictated stuff (50+ words, stutters, "um", no punctuation).
- agent-model: picks the model for subagents spawned without one. Sonnet for simple things (lookups, mechanical edits), Opus when it needs stornger reasoning. If it can't tell, it asks me.
- at-a-glance (a bit of a melting pot): 1 line above the prompt with plan usage, running subagents, my open PRs, new Slack mentions, and a "Braver" button. Braver tells Claude to go for the root-cause fix instead of the cautious patch, for one prompt only. Click the line for the detail pane.
Adrien retweeted
(Tonight) we're opening up two company brains, live. ๐ง
Yochan Khoi shows how Gorgias built Cortex in-house. @Christophepas shows what Slite looks like. You'll see the same four components in both, solved two different ways.
Half the hour is live demo, and we're keeping real time for Q&A at the end, so bring your questions.
Join us, at your local time: 7PM CEST / 1PM ET / 10AM PT
Last chance to save your spot: app.livestorm.co/slite/compaโฆ
Adrien retweeted
Internal marketing matters!
We shipped a visual generator for our gtm, it's called "Love at first Slite"
:chef_kiss:
Adrien retweeted
"Langfuse allowed us to go from reliable changes every few months to iterating multiple times a week..."
@slitehq builds an AI knowledge base that keeps a company's docs current. And they evaluate every change with Langfuse datasets!
langfuse.com/users/slite
I've seen the slides and the talk about company brain is gonna be worth it๐ฅ
Will be at @TechBBQ in CPH for the rest of the week ๐ฉ๐ฐ!
Ping me if you're there and come see my talk on Company Brain ๐
did a little tool to find out what your coding agent actually sends to the model: the full system prompt, tool, etc. none of it is in the session files.
tiny local proxy + viewer.
works with Claude Code and Codex, everything stays on your machine.
github.com/adrientaravant/agโฆ
Adrien retweeted
One creator on your team is nice. Five is a distribution channel.
(and it's the only one your competitors can't outbid you for)
In July, I looked at my own LinkedIn recap of last year: 4.3M impressions, $0 on ads.
That's one person. We've now got a whole team doing it, across platforms.
Why Employee Generated Content (EGC) is still worth starting now:
๐ญ. ๐ฃ๐ฒ๐ผ๐ฝ๐น๐ฒ ๐ณ๐ผ๐น๐น๐ผ๐ ๐ฝ๐ฒ๐ผ๐ฝ๐น๐ฒ
Nobody don't follow a logo to learn something, because they know they're being sold at. A company account has one voice, one niche, one product. Your team (founders, designers, engineers, etc) each cover different niches, networks, formats, and will reach people who would never have found the brand.
๐ฏ. ๐๐ ๐ฐ๐ผ๐บ๐ฝ๐ผ๐๐ป๐ฑ๐, ๐ฎ๐ฑ๐ ๐ฑ๐ผ๐ป'๐
Paid stops the second you stop paying. People are still engaging with posts we shared months ago. Now run that across a whole team.
๐ฐ. ๐๐ ๐ฐ๐ฎ๐ป'๐ ๐ฏ๐ฒ ๐ฐ๐ผ๐ฝ๐ถ๐ฒ๐ฑ
A competitor can probably match your ad spend by end of week. But they can't match your team's credibility, and they definitely can't buy their way into being trusted.
This dashboard is an internal diamond for our LinkedIn insights built by Adrien Taravant. Go follow him for AI and GTM insights, he's brilliant!
And the rest of the team: @Christophepas, Nils Jacobsen, @pierrerenaudin, Katerina Alexaki, Cameron Pemberton and @_ishaanxgupta. Founders, engineers, design, sales. They were not hired to post, but they just have ideas worth sharing. ๐ซถ
Is your team posting? And if not, what's stopping them? ๐
I knew u could accelerate so much your agent with the proper search service like exa.ai, today really felt the difference for "historical research"
2 min 57s later: "I reviewed 169 search results across five angles, with substantial duplication, then checked the primary text, German dictionaries, academic histories, and current reporting."
Adrien retweeted
As someone tired of reading AI slop documents all the time love this policy from Clay
linkedin.com/posts/vaanand_wโฆ
Adrien retweeted
Why is graph engineering suddenly everywhere, weren't we just talking about loops?
Here are the 6 terms you need to know to keep up:
(Plot twist: a loop is just one node inside the graph.)
Graphs are hot right now for one reason, loops alone got flaky. An agent that just acts โ checks โ repeats is powerful, but you can't reason about where it'll go. A graph gives you the map: fixed states, clear routes, and checks you control.
๐ญ. ๐ก๐ผ๐ฑ๐ฒ
One step in the graph. It can hold a single tool call, an agent, or a whole loop that works something out before handing the result off. The unit of work.
๐ฎ. ๐๐ฑ๐ด๐ฒ / ๐ฅ๐ผ๐๐๐ฒ
The paths you draw between nodes. One node can have several outgoing routes, the work only travels one of them.
๐ฏ. ๐๐ผ๐ป๐ฑ๐ถ๐๐ถ๐ผ๐ป๐ฎ๐น ๐๐ฑ๐ด๐ฒ
The decision point. It reads the current state (with code logic or an LLM) and picks where the work goes next. This is where your quality gates and safety checks live.
๐ฐ. ๐ฆ๐๐ฎ๐๐ฒ ๐ ๐ฎ๐ฐ๐ต๐ถ๐ป๐ฒ
The shape of the whole thing. Every node is a state the work can be in, and a check at each one decides the next hop. Built a flow in n8n? You already drew one.
๐ฑ. ๐ฆ๐ต๐ฎ๐ฟ๐ฒ๐ฑ ๐ฆ๐๐ฎ๐๐ฒ
One object that travels the entire graph; every node reads it, updates it, hands it on. The state machine is the ๐ด๐ฉ๐ข๐ฑ๐ฆ; shared state is what actually moves through it. { messages, context, next }
๐ฒ. ๐๐ด๐ฒ๐ป๐ ๐๐ผ๐ผ๐ฝ๐
The autonomous part. The agent acts, evaluates the result, and repeats, no human poking it at every step. It runs until the task is done.
So loops didn't die. They moved ๐ช๐ฏ๐ด๐ช๐ฅ๐ฆ the graph.
Loops give your agent freedom.
Graphs give you control.
The good systems in 2026 use both, a loop where you want autonomy, a graph around it for everything you can't afford to get wrong.
Loops or graphs, which are you reaching for more right now? ๐ค
Adrien retweeted
The company brain hype left X.
126 people told me where it went, and why:
Some weeks ago I shared a survey around the company brain, and asked you guys if you're already building or using one.
Here's some tiny insights I got already (more in the final report later):
/1
The hype left the X / Twitter bubble - most of you didn't hear about the company brain from the tweets that kicked it off. ~60% found it secondhand (articles, videos, word of mouth.)
/2
People want it for themselves, not just for the AI - majority of you guys want a company brain for your own team, not just to feed your agents.
/3
The context problem is nearly universal - almost everyone is drowning in scattered, fragmented context. And single agents + docs haven't fixed it yet..
Didn't expect the responses to be this good. Thank you to everyone who filled it in. ๐ซถ
If you haven't filled it in yet, this is your last chance. ๐
Takes no more than 4 min, and get a chance to win 200$ gift card + a quote in the ebook: docs.google.com/forms/d/e/1Fโฆ
If youโre building a company brain, GBrain is required reading.
Its 6 layers explain the machinery and where the company problem begins.
GBrain is Garry Tanโs open-source brain layer for his AI agents. It ingests meetings, emails, voice calls and notes, connects them, and gives his agents memory beyond one chat.
Basically, one layer feeds the next.
1/ The knowledge graph turns flat notes into entities and relationships.
2/ MCP connects the agent to the external tools where information lives.
3/ Skills tell the agent what to do with those tools.
4/ Hybrid search combines semantic meaning with exact keyword matching to find the right material.
5/ Agentic RAG decides how to retrieve that material, checks it, and uses it to build an answer.
6/ A company brain adds the maintenance layer that keeps the whole system current, trusted and usable.
My take: GBrain is a very good blueprint for a personal brain.
The company brain problem gets quite a bit harder.
Multiple people bring different permissions, conflicting sources, stale docs, and updates that need clear ownership/approval.
That is the gap between personal memory and shared company truth.
Still, GBrain is one of the best repos to play with if you want to understand how the plumbing works.
Adrien retweeted
2.5m impressions on a KB architecture sounds absurd.
Then you learn it serves 700+ people and their AI agents, answering 15,000 questions/day.
And the crazy part is they ditched Karpathy and Garry Tan's context graphs.
Here's what to copy, what to skip, and how to have it live this week:
Adrien retweeted
We're hiring a Technical Product Manager at Slite.
You'll work directly with @Christophepas (CEO) and @pierrerenaudin (CTO), embedded in a ~10-person engineering team.
What you'll do:
โ Co-own the roadmap
โ Make hard calls on scope
โ Shape how knowledge stays alive in the agentic era
Part strategist, part engineer, part product visionary.
If you've shipped products at scale, can read a PR, and speak LLMs, embeddings, and agents fluently, we'd love to talk.
Apply here: slite.com/jobs/technical-proโฆ
