Watch Desk posted an update
Uber exceeded its AI budget early this year, then tightened the defaults on its coding tools, VentureBeat reports. The changes include automatic context compaction at 400,000 tokens and medium reasoning effort.
Why it mattersUber engineers also say the company now uses an internal context graph to give coding agents targeted system maps, instead of supplying large amounts of documentation up front. Less context can mean less spend, though the report does not quantify the savings. It is a useful reminder that controlling AI costs can mean changing how tools are configured, not just picking a cheaper model.
Discuss: Would you prefer coding agents to start with lean context and ask for more, or receive the full documentation from the outset?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.