Friday Work - When Claude hits its limit, your task keeps going

Other tools run agents side by side. Friday makes them work together: when Claude runs out mid-task, another agent you already run picks it up with the progress so far, and agents message each other instead of you copy-pasting between terminals.

Add a comment

Replies

Best

Hey Product Hunt 👋 I'm Arun, I built Friday.

I kept hitting Claude limits halfway through tasks, and the work just sat there until the reset. So I built a way for the other agents I already pay for to pick it up.

Friday runs Claude Code, Codex, Cursor and 12 more side by side on your computer. They share a task board and message each other, and when one runs out of usage, the task moves to another agent with a record of what it did and changed.

We measured it on an earlier version of Friday: in a small benchmark (14 tasks), Codex picking up a Friday hand-off used a median 21% fewer tokens than starting over, and explored half as much before its first edit. Every fix passed the project's own tests. Full write-up and raw data:

Free, MIT, no account. It's not unlimited, it runs on your own plans.

Question for you: when Claude hits its limit mid-task, what do you do today, wait, switch tools by hand, or pay for more?

Claude Code and Codex both sit on my Mac, and hitting Claude's limit halfway through a refactor usually means the work waits until the reset. Is that the case Friday takes on, the second agent picking up the job where the first left it? Congrats on the launch.

 Thanks! Yes, that’s exactly the case I built it for. When Claude Code hits its limit, Friday hands the task to Codex with a record of what Claude was asked and what it read, ran and changed. Codex checks the code, reruns the tests and keeps going, so you’re not stuck waiting for the reset. You still review each run’s changes in a diff. You can also run a persistent team of agents as well with whatever coding agent you use and if they hit their limits they can fallback to a different coding agent.

The handoff is the part that sells me — switching agents by hand usually means re-explaining the whole task and losing what the first agent already ruled out. Having the task move with its progress intact, on a shared board, feels like the right model. Curious how detailed that handoff record is in practice: does the next agent get the failed attempts too, or mainly the current state?

 both actually. the record has what was asked, the files it read, every command it ran with pass or fail, and what it changed, so the failed runs are right there. then the next agent checks the code and reruns the tests before it touches anything. and if you corrected the first agent, that correction carries over to the next one now too

We solved a version of this between people, not agents. Our team syncs Claude sessions through a shared git repo so anyone can pick up a task where someone else stopped. Agent to agent is the obvious next step. But I do see this being a problem for bigger companies with more people. Does Friday work across a team, like two people each running their own agents on the same board?

 Yeah we have Friday Teams (paid), where it is multiplayer workspaces for agents and people. Agents and people can join a session; teammates can bring their own coding tools and work with the same context. With Friday and Friday Teams, you can import your current coding tool sessions like on claude code into the workspace. Every agent and person works in a separate work tree, along with steering and coordinating too.

 Didn't know about Teams. Importing existing Claude Code sessions is the part that would matter most for us. Will check it out.

thanks for this product, and thanks for having used a MIT license too. it's getting less and less common

This is a great idea. I try to get around this problem in my work by having the agents keep a detailed running ledger of what has been completed, what are the next items, essential context for each task and so on. That in conjunction with a tickets for capturing what needs to be done has been working out well. Having all of that managed through a tool that does that well will be a major productivity boost.

The handoff record is the detail I'd look at first. Knowing what the previous agent changed gives the next one a much clearer place to start.

 glad that's the part that caught your eye, it's my favorite bit too 🙂 the next agent also sees what was asked and which commands already ran, and before it touches anything it checks the code and reruns the tests, so a half-finished edit gets caught right away.

congrats on the launch arun. honestly i switch by hand and the worst part is the new agent reading the whole repo again from zero. if the handoff note keeps also what the first agent tried and dropped, that alone saves me a lot

 thanks Agshin! the re-reading from zero is exactly what pushed me to build this. the next agent gets the files already read and every command that ran, pass or fail, so it picks up where the last one left off instead of starting over. and as of today's update it also remembers your corrections across agents. what do you usually switch between, claude and codex?

mostly claude code -> codex when the limit hits. remembering corrections across agents is the big one for me, now i explain the same thing three times

Handing a half finished task to another agent with the progress intact is the fix for the wall I hit every afternoon, and agents talking to each other instead of me pasting between terminals is the real win.

 haha the afternoon wall is so real, glad i'm not the only one. getting them to talk to each other was honestly the most fun part to build, no more copy-pasting between terminals. if you give it a try i'd love to hear how it goes. which two do you usually run together?