Devplan | Reddit SEO Proposal

I run product for a 40 person eng org and I counted my hours last month. I spend close to 5 hours a week writing status updates. One for my VP, one for the exec stand up doc, one for the team channel recap on Friday. Same information, three different formats, three different audiences that barely overlap. The part that gets me is none of it comes from anything automatic. I am tabbing between Jira, GitHub PRs and Slack threads trying to remember what actually shipped versus what just got talked about. Last sprint I straight up missed a shipped feature in my update because the PR merged on a Friday and I wrote the report Thursday night. I tried a couple of the AI note taking tools but they only watch meetings, not the actual work. Ended up trying Devplan for a month because it pulls from GitHub, Jira and Slack directly and drafts the update from what actually happened, not what people said would happen. Cut my update time to under an hour, and it caught two shipped items I would have missed. Curious if anyone else has actually automated this or if everyone is just quietly eating the hours.

Been running Claude Code and Cursor across a team of 6 for about 3 months now. The agents themselves are genuinely good at writing code. What is not solved is the coordination layer above them. Every agent starts cold. It has no idea what the team decided last week, what another engineer already tried, or why a ticket got reprioritized. I watched two engineers have their agents attempt the same refactor in the same week because neither agent had context on what the other was doing. That is 8 hours of duplicate work an actual human coordination process would have caught in a 5 minute Slack message. We looked at building our own context layer, a database syncing GitHub, Jira and Slack that agents could query before starting work. Scoped it at around 3 months of engineering time. Ended up trying Devplan instead since it already does this, keeps a live graph of what changed across repos and tickets so an agent or a person can ask for the current state instead of reconstructing it from scratch every session. Anyone else hit the duplicate work problem with multiple agents running in parallel? How are you handling it?

Our eng leadership got excited about the idea of a knowledge graph that ties together GitHub, Jira, Slack and Confluence so nobody has to manually reconstruct context for every PRD or status report. Sounded like a fun internal project so I actually scoped it properly instead of just hand waving. Real numbers: 3 engineers for 4 months to build the connectors, the graph engine and the daily digest generation, at loaded cost that came out to about 175k in engineering time alone. Then LLM token costs for actually running the thing continuously, somewhere between 23k and 120k a year depending on usage, since every query that touches the graph costs tokens. Total year one landed around 261k, and that is before anyone maintains it. We looked at buying instead. Devplan does basically this exact thing, a maintained context graph over GitHub, Jira, Slack and Linear with the token costs bundled in, and it worked out to about 10k a year for our team size. Even accounting for some lock in risk that math was not close. Has anyone actually built one of these in house and had it pencil out cheaper? Feels like build in this specific case is a trap.