srbsa

I analysed 10K open-source PRs for something I'm building and found that anywhere between 42% and 67% of review friction is due to context beyond the codebase.

From a workflow PoV, how annoying is the manual context synthesis to curate the prompt/agents.md? Is review a solved problem for teams with review agents, or is it now the real blocking step?

From what I've seen it sits at three levels:

Layer 1: Personal (memory MCP tools, personal *.md files) Layer 2: Repo (CLAUDE or AGENTS.md, in-repo docs) Layer 3: Org-wide (Unblocked, Glean, or similar)

Most people have strong opinions on 1 and 2. Layer 3 seems underinvested.

Curious, do you think 1 and 2 make 3 redundant, or would a strong 3 mean you'd stop/reduce patching with 1 and 2?

I've been talking to engineers, mostly in large teams. While they love cross-surface search with Glean, they still assimilate and curate the context for agents manually. It is especially useful during incidents and new onboarding.

I've been building a knowledge substrate for engineering teams that is native to coding agents and pulls from all connected tools. It returns cross-source, fresh, credible evidence in agent-native form without context bloat. It's effective and efficient enough to deploy for smaller teams.

My ICP is growing teams/startups between 15 and 80 engineers with good agentic coding adoption. I've been trying to reach out to these founders/teams. A very low number of conversations materialise, and the ones that do perceive it as only 'good to have'.

Have I been building vitamin?