Why coding agents need shared context

By Arun Dass · Published 3 October 2026 · Updated 2 November 2026

Coding agents share context when the second one starts from a record of what the first was asked, found and changed, and can read the same messages and task board. In Friday, the free open-source app, that happens in one session on your own computer. Without it, every agent starts knowing only what you type.

What is context, for a coding agent?

Context is what the agent has learned about the task: what you asked, what it read, what it tried, what failed and why, and what you decided along the way. It is the difference between “fix the login bug” and “fix the login bug, which is not in the session code, we already ruled that out”.

Most of it lives in one conversation, in one tool.

What goes wrong when context is not shared?

  • You switch from Claude Code to Codex and explain the task again from scratch.
  • A second agent repeats the investigation the first one finished an hour ago.
  • Two agents make different assumptions about the same code, because neither saw the other’s findings.
  • You cannot review a change without reading a whole private transcript, or guessing.

Each of these costs time, and each also costs tokens, because an agent that starts cold re-reads and re-tries what the last one already did.

What does it look like when it is shared?

SituationWithout shared contextWith agents that share it
Switching from one coding tool to anotherRe-explain the taskThe task moves to another agent with a record of what the first did and changed
Two agents on one jobEach guesses what the other changedThey message each other, for example about a rename, and a question gets an answer
Splitting a goalYou write each prompt by handOne agent posts a task board; the others pick up the tasks that suit them
Reviewing what happenedDig through a private chatEvery message, command and file change in the session, and a change set per run

How does Friday share context?

Friday runs your coding tools side by side in one session on your own computer. Agents share a task board: one agent splits a job into tasks, each naming the files it owns and what it waits on, the others pick up the tasks that suit them, and each finished task is checked before it counts as done. Agents also message each other directly, a heads-up about a rename or a question one of them waits on.

When a tool hits its usage limit, its task moves to the agent best placed to finish it, with a record of what the first did and changed, instead of starting over. The record is built from the session log. It does not include the full conversation or anything the first agent never wrote down.

The agents run on your own plans and sign-ins. Friday has no account, and sends anonymous usage counts you can turn off, never your code or prompts. The free open-source app runs on macOS, Windows and Linux.

What should you still pass on yourself?

Shared context does not replace a clear instruction. Say what “done” means, name the files that matter, and write down any decision an agent could not know. A short note at the start of a task is worth more than a long transcript.

Common questions

Is shared context the same as a bigger context window?

No. A larger window lets one agent hold more. Shared context is about the same information reaching the next agent, instead of staying in one conversation.

Do different AI models see the same context in Friday?

They share a task board and messages in the same session. Each agent keeps its own way of reading them, so a clear instruction at the start of each task still helps.

Does sharing context send my code to Friday?

No. Friday runs on your own computer with no account, listening on 127.0.0.1 only. It sends anonymous usage counts you can turn off, never your code or prompts. Your coding tools talk to their own providers as they always do.

Run your next task in Friday.

Free and open source (MIT), on your own computer: the desktop app for macOS, Windows and Linux, or one command with Node.js 22.