article
Most AI pilots should be killed sooner
The polite fiction around enterprise AI is that every pilot teaches something. Many teach only that nobody defined the decision, owner or failure cost.
Emil Shirokikh · Published September 21, 2026 · Updated September 23, 2026 · 2 min read
Abstract
AI pilots survive because ambiguity is politically convenient. A demo can look impressive while avoiding the harder question: which operating decision changes if the system works?
The kill test
A pilot needs one named decision, a current baseline and a date when evidence will be judged. Without those, activity can continue indefinitely without becoming useful.
Practical test
Write a one-page test charter: decision changed, accountable owner, baseline, target, prohibited failure and retirement date. If the sponsor will not sign it, stop.
Evidence to retain
The output is a signed go, change or stop decision—not another demonstration.
Challenge the conclusion
Exploration can be valuable when uncertainty is explicit. Record what will be learned, how it will be observed and which result ends the experiment.
Use this in a working session
Bring the sponsor, operator and one independent reviewer. Leave with a named owner, a dated decision and an inspectable artefact—not another presentation.
BELTO editorial analysis. It does not describe a client engagement or claim a commercial result.
References
2 sourcesAuthor
Emil ShirokikhFounder
Founder of Belto Inc. Writes on engineering, venture building and applied intelligence.
Related
Read next
AI agents need boundaries, not personalities
The market is decorating automation with human traits while neglecting permissions, reversibility and evidence.
A computer-vision demo is not a system
Benchmark accuracy says little about glare, drift, maintenance, latency or the operator who must act on an alert.
Private AI is an architecture choice, not a product
Buying a server does not make AI private. Data paths, administrative access, telemetry and update channels still decide the boundary.