A Microsoft Research / Stanford study (arXiv:2604.22750, April 2026) found agentic coding tasks consume ~1,000× more tokens than chat on the same model. Input dominates the cost, not output. Accuracy gains can saturate; the dollar cost depends on rates and caching. Here is what that means for the input surfaces you can actually compress.
On July 13 the Pentagon suspended CMMC Phase II, the third-party audit that was set to start in November. NIST 800-171 self-assessment stays, and so does the legal duty to protect federal data. For contractors running AI dev tools, the burden of proof just moved onto them.
Engineering, research, cost, and tutorial posts in one feed.
Long-form pieces from the team building gotcontext. Compression internals, model evals, customer post-mortems. RSS feed available — see /blog/rss.xml. Unsubscribe with one click.