QSTVE / Reading path
AI changing work
How do we tell whether AI is making work better?
Begin with the job and a way to check it. Compare useful results, the cost of review and the limits of the evidence before choosing a model or expanding a workflow.
Choose the route after the job
In Man and Machine, Dave Glaser describes a deterministic task that still called a premium model. The work finished, but the model added nothing. His resulting rule is to begin with the need for judgement, structure, privacy or speed, then choose the smallest route that can do the work well. Some tasks need no model.
Follow improvements into the next tools
The Next Experiment Has Changed follows dated reports of AI improving research software and computing infrastructure. Its distinction matters: a faster calculation can change what researchers can explore, but does not establish faster whole research programmes or a dependable exponential growth rate. Developer reports and preprints retain their stated limits.
Close the improvement loop with an outside check
The accepted Field Guide for When AI Starts Rewriting the Company recommends fixing the objective and protected outcomes, freezing a baseline, limiting the change surface and using independent evaluation. A narrow canary and a demonstrated rollback turn a proposal into a bounded test. This is operating guidance, not proof that autonomous improvement is production-safe.
Read the argument
Supporting essays
The Most Expensive Token Is the One That Did No Useful Work
A year of personal AI produced a simple buying rule: match the route to the work, and use no model when none is needed.
The Next Experiment Has Changed
AI Work Needs Proof, Not Just a Good Explanation
If you cannot do the work yourself, you still need a clear way to check that it was done right.
Put it to work
Practical Field Guides
When AI Starts Rewriting the Company
Agents can study outcomes and propose better procedures. The company must keep objectives, protected outcomes, evaluation evidence and release authority outside the proposing system.
AI Work Needs Proof, Not Just a Good Explanation
Choose the proof before an agent begins. Verify the changed result, follow evidence for important claims and measure the time spent reviewing and repairing the work.
Nine Crons Were Running. I Thought It Was Four.
Configuration records can drift from running agents. Compare the actual jobs, model and access settings with the approved configuration, and retain the differences.