Define the win.
Decide what acceptable means before seeing a set of alternatives.
A WORK CONSIDERED FRAMEWORK · AUGUST 2026
Generative AI is not a casino. But left unguided, model choice can feel alarmingly like one. The Knowledge Casino turns novelty, retries and hidden cost into a calmer system for getting to a verified answer.
01 / THE PROPOSITION
Decide what acceptable means before seeing a set of alternatives.
Find the least costly model that reliably clears the quality and risk threshold.
Once acceptance and verification are complete, stop generating alternatives.
02 / THE MODEL PORTFOLIO
A small portfolio reduces decision fatigue and makes performance learnable. The question is not “which model is best?” It is “what gives this task the best odds of a verified acceptable result?”
For drafting, analysis, synthesis and most mixed knowledge work. The default should earn its place.
03 / THE OPERATING LOOP
Seven moves that turn a flashing model picker into a small, explicit decision system. Scale control to consequence.
Define the job and the finish line.
Score consequence, reach, persistence and difficulty of checking.
Choose the smallest machine likely to clear the bar.
Keep conditions steady enough to learn.
Measure first-pass acceptance and repair minutes.
Verify the claims or actions that matter if wrong.
Accept, make one targeted retry, change approach—or stop.
04 / HOUSE RULES
Behavioural guardrails for the moment a human being is staring at a menu of machines.
05 / ASSESS THE STAKES
A low-stakes brainstorm should feel light. A consequential recommendation should not be allowed to walk past the evidence desk.
Light checks. Creative variation can be welcome.
Check key claims. One targeted retry.
Source-level checks. Diagnose, don’t spin.
Independent validation and human approval.
06 / START HERE
Use it individually, make it a team practice, or begin with a one-page decision card. The full paper includes task passports, model bake-offs, a controlled pull log and the Shiny Machine Gate.
Twenty-one sections, worksheets and worked scenarios.
READ / DOWNLOAD PDF ↗ONE-PAGE TOOLTen questions to ask before you pull the lever.
START THE DECISION TREE ↗TEAM PRACTICECompare options on representative work, not launch-day theatre.
OPEN RESEARCH PAPER ↗WORK CONSIDERED
Technology should not only be judged by what it enables organisations to do. It should also be judged by what it asks people to become.RETURN TO THE TABLE ↑