The Developer’s Guide to the Agentic Git Forge: Agents That Remember

Coding agents have changed how we write software, but the tools around them were built for humans working at human speed. Git forges were designed to store code and review diffs, not to hold the reasoning, context, and decisions behind thousands of agent-generated changes. This workshop looks at what a forge built for both agents and developers should look like, starting with the foundation: memory. Today’s coding agents start every session with amnesia, and teams pay for it in three ways:

Context disappears when the session ends - The reasoning behind your code, including decisions, constraints, and “we tried that, it broke prod,” is scattered across sessions that the next agent never sees. Agents burn tokens rediscovering your codebase - Before writing a line, every fresh agent spends its budget re-learning structure and conventions the last one already knew. Stale memory erodes trust - When code changes but remembered facts don’t, agents act confidently on outdated information, and reviewers can’t tell what to trust.