Pinned Tweet
we are a two person software agency and agents write almost all our code now.
what actually works, and what bit us 🧵
you can now shrink logs and tool output before your agent reads them, headroomlabs-ai/headroom
one headroom wrap claude, and their 10k token log goes to 1.2k with the same FATAL found
prose and dense output barely shrink though...
Graphify-Labs/graphify lets your agent query a graph of the repo instead of grepping files
before a refactor, graphify explain shows everything linked to a class, read or inferred
how does your agent find its way in a big repo today?
DietrichGebert/ponytail makes your agent reuse whats in the repo before writing new code
their date picker demo just reuses Input and the browser picker, 10 lines instead of 335
all the numbers are from their own benchmark though...
nobody opens the time tracker during the day. the hours get typed in later, from memory, and some never make it in
now a consultant can type "log 3 hours on severn yesterday, code review" into claude code and the entry is booked
works in claude code, codex, cursor, vs code and any app that can add an mcp server. each person uses their own key, with the same rights they have in the app
consultingcockpit.com/featur…




















