TLDRocket
Sign in

Prompt tips from Anthropic

Ben's Bites

Keep two files for every agent: chat-index.md and folder-map.md. It sounds boring, but it may beat constantly resuming old sessions.

Based on reporting by Ben's Bites — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

Ben’s Bites packed a lot into one issue, but the most practical tip came near the end: keep two simple files that any agent can read. One is chat-index.md, a daily-updated map of past conversations. The other is folder-map.md, which explains where each project lives. Instead of trying to drag an old thread back to life, the idea is to start fresh and point the agent at the right history and the right place.

Keshav says the workflow is straightforward. Open a new session, then tell the agent: read the chat where I was working on a specific project, and we’ll continue from there. That turns the file system into memory. No elaborate setup, no special product required, just a couple of text files and an automation to keep one of them current.

The rest of the newsletter is a reminder that the agent boom is moving in a thousand directions at once. Google’s Gemini 4 Argon is out for everyone to use, though Ben found it slow and unimpressive for design work. Claude Code’s desktop app picked up a design refresh, multi-folder environments, child sessions, side questions in the /btw tab, and even its own version control system instead of git. Grok Bot 0.70.2 added live cards for Figma, Notion, and Canva, plus X search without account linking and checks before irreversible actions.

There were also signs that the tools are getting more opinionated about workflows. Attio is reworking itself to be agent-first. Ramp’s Glass is built around shared AI workspace and company memory. Factory says its droids can run on your own machine or your own cloud. And one line from the issue cuts through the noise: giving coding agents a docs link doubled task completion from 41% to 84%. The message is pretty clear. Context beats vibes, every time.

My take — AI-written commentary, not fact-checked reporting

The industry keeps selling miracle agents, but the real trick is still housekeeping. A couple of plain text files beat another shiny chat box with amnesia. That’s not glamorous, which is probably why it works.

Read more about this at: Ben's Bites

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.