Before deleting the generated tests, find out what they protect
Three tidy tests can repeat one weak check while an awkward older file catches the mistake that matters. Clean up around the behavior people still rely on.
AIOps
AI in production: tools, operating patterns, costs, and failure modes.
The Codex worker may still be running after the window closes
A host restarts while a coding worker is making a patch. Recovering the process, its files and its permissions is the difference between continuing the job and doing it twice.
Four long investigations, one Blackwell workstation: trying Basalt
Basalt makes specific choices about where a large model lives. Follow four hypothetical users to see how those choices affect memory, waiting and useful answers.
Which balance pays for your Claude agent?
A backup-review job can run the same prompt and draw on a different allowance. Claude’s October API-credit update makes the launch path, sign-in and finished task worth following together.
Sharing one model between an AMD card and an NVIDIA card
Two cards may make room for a model that fits on neither alone. Infermeld exposes the layer choices, but each card still has to hold its own work when the real request arrives.
Instinct: when the conversation becomes a booking
Finding a handyman is useful. Reserving the afternoon changes someone else’s calendar. Follow the difference through permission, a dropped call and the provider’s confirmation.
Kimi K3: keeping the small clue in a million-token investigation
A timestamp buried in a large incident packet can overturn an easy explanation. Kimi K3 offers more room for the record; its value depends on the connection the answer makes.
ChatGPT Space: helping the next shift pick up the incident
The rollback is complete, but orders are still waiting. A shared page can carry that unfinished investigation into the next shift without making the story sound resolved.
ninfer-ext EXL3: what the smaller file buys when generation slows
The smaller artifact saves 1.44 GiB but generates more slowly in the project’s measurements. Its value depends on whether that space makes a required request possible.
DeepSeek Harness: which workspace is behind the desktop window?
The same request can read a laptop file or a remote copy. Follow a README question through Harness profiles, plugins and model routes to understand the answer it returns.
Pi.dev: a coding agent that can grow around your work
Start with a small bug, a repository question or a source-backed document. Pi supplies the terminal agent; templates, skills and extensions let you change the parts you keep needing.
Jev can pick a support queue. The ticket still has to get there.
A typed answer saves an application from parsing a paragraph. Follow a hypothetical support ticket to see what Jev chooses, what its probabilities mean, and who still has to deliver the work.