DEV Community: rohanprichard

TalkToMe: I made my AI agent call me.

rohanprichard — Tue, 29 Sep 2026 03:00:32 +0000

I built TalkToMe — an open-source Mac app that lets your coding agent call you from the same session it’s already working in. Try it out and let me know what you think!

rohanprichard / talktome

Voice calls with the coding agent you already use. A macOS menu-bar app.

talktome

TalkToMe is a macOS menu-bar app for voice calls with a coding agent The agent rings you from the session that it already runs. You answer, and you talk.

The agent keeps its model, tools, files, and history. TalkToMe does not supply a language model The app records the microphone, turns speech into text, sends the text to the agent, and speaks the reply.

How a call works

You ask the agent in its session to call you, for example "Call me with TalkToMe."
The agent runs talktome call with its session ID.
TalkToMe shows a ring on the screen. You select Answer.
The agent speaks a short greeting.
You speak. The agent receives your words as a message in its session and replies.
You select End, or the agent runs talktome end.

The app has no window that starts a conversation After setup, TalkToMe stays…

View on GitHub

I tested PRAXIST by Sapient Intelligence

rohanprichard — Sun, 30 Aug 2026 10:19:40 +0000

Praxist is a new autonomous R&D tool that gives multiple AI coding agents a shared experiment loop: they can test ideas, evaluate results, keep track of failures, and use earlier findings to decide what to try next.

I tested it on the Adult Income dataset rather than a toy demo. It found a small repeatable feature-engineering improvement, rejected several worse options, and left behind a record of the experiments.

Full write-up:
https://medium.com/@rohanprichard/i-tried-praxist-by-sapient-labs-on-an-ml-dataset-ce6baecfa6ae

I was maxing my Claude 5-hour limit daily and still wasting weekly quota every night, so I built a tool that spends it while I sleep

rohanprichard — Mon, 27 Jul 2026 18:23:40 +0000

Like a lot of you I hit the 5-hour cap most days. What actually annoyed me was realizing the weekly limit doesn't line up with that. Even capping out daily, I ended every week with quota unused. It expires overnight even after I paid for it.

So I built claude-overnight. I queue questions during the day, /queue how do sqlite WAL checkpoints work? right inside Claude Code, and a scheduler runs them at night once my limits reset, through claude -p on the subscription. Morning brings markdown reports and a digest of what ran and what happened.

Every job saves its claude session, so overnight resume <id> reopens the conversation that wrote the report. You can argue with it about its conclusions over coffee. Or overnight followup <id> "go deeper on X" and it continues tomorrow night.

Coding tasks work too. They run in a throwaway git worktree on an overnight/* branch, only against repos I've explicitly trusted, so the agent never touches my working tree. Morning review is just git diff main..overnight/whatever.

Since people will ask how it reads limits when there's no official API: Claude Code stores an OAuth token locally (Keychain on Mac, ~/.claude/.credentials.json elsewhere), and GET https://api.anthropic.com/api/oauth/usage with that token plus an anthropic-beta: oauth-2025-04-20 header returns your 5h and weekly utilization with reset times. Same trick the menubar trackers use. It's undocumented and the response shape already changed once while I was building this, so the tool survives without it.

The design constraint I cared most about: don't eat my own morning quota. It won't start above 20% of the 5h window, stops at 60%, skips entirely past 80% weekly, rechecks between jobs.

In the morning it opens a page in the browser with the whole batch on it — what ran, how long it took, the resume command for each one, and every report rendered inline so you're not clicking through files half-awake.

Check it out at https://github.com/rohanprichard/claude-overnight

Curious what you'd queue overnight, honestly. Let me know!

AI Can’t Learn. It’s Just Really Good at Pretending.

rohanprichard — Sat, 18 Oct 2025 18:36:39 +0000

Sounds controversial? See what benchmarks are measuring for yourself.

Every week we hear about frontier models crushing human tests and breaking records. But here's what nobody's talking about: these models can't learn anything new.

Check out what I think about the learning problem in AI and where this race is actually going to lead.