Pillar guides
A buyer’s guide to AI agents for operations
How to tell an agent project that will reach production from one that will not — before you spend anything. Scoping, integration feasibility, evaluation, and the questions that separate engineering from sales.
In this guide
- What an agent is, and what it is not
- The five questions to ask any vendor
- Why integration decides feasibility, not the model
- Reading a fixed price versus an estimate
- What a good scope document contains
[FULL GUIDE — TO BE WRITTEN]
Evaluating AI agents: the discipline nobody sells you
The single largest difference between an agent that holds up and one that quietly degrades. What an eval suite is, how to build one from your own cases, and what to require contractually.
In this guide
- Building a regression set from real cases
- What to measure, and what not to
- Detecting drift three months after launch
- Running evals through a model deprecation
- What to put in the contract
[FULL GUIDE — TO BE WRITTEN]
Integration reality: what your systems will actually allow
The audit’s hardest question, answered generically. Which platforms expose usable APIs, which are partner-gated, which need a bridge, and how to test before committing.
In this guide
- Read access versus write access
- Partner-gated APIs and how to get through them
- Legacy systems without APIs
- Rate limits, pagination and partial failure
- Testing feasibility in five days
[FULL GUIDE — TO BE WRITTEN]
From pilot to production
Why most agent proofs of concept never deploy, and the sequence that avoids it. Shadow mode, staged rollout, rollback paths, and the escalation design that earns trust.
In this guide
- Scoping for deployment, not demonstration
- What shadow mode proves
- Staged rollout: overflow, percentage, primary
- Designing escalation
- The rollback path you test before you need it
[FULL GUIDE — TO BE WRITTEN]
Agents in regulated and field operations
What changes when the work happens on a site, under a permit, with intermittent connectivity and a regulator. Evidence capture, attestation boundaries and offline design.
In this guide
- Why the agent never attests
- Contemporaneous capture versus reconstruction
- Offline-first field design
- Modelling overlapping obligations
- What auditors actually want to see
[FULL GUIDE — TO BE WRITTEN]