Every AI L&D consultant currently has a demo. Mine included. The demo is the least useful part of the engagement, and if it arrives in the first meeting you are being sold a tool rather than a decision.

The job is not to show you what the technology can do. You already know. The job is to work out which parts of your operation should change, in what order, and what you should refuse.

That distinction decides whether your team ends up more capable or simply faster at producing the wrong things.

What should an AI L&D consultant do first?

Map where your time actually goes. Not the process diagram, the real hours: scripting, reviews, rework, chasing sign-off, rebuilding the same module for a third audience.

A good AI L&D consultant starts with your bottleneck, not their stack. Most L&D teams lose more time to unclear briefs and slow approvals than to authoring, and no model fixes an approval culture.

For example, a team asks for generative drafting when the real constraint is six reviewers and no agreed standard. Fix the standard first and the automation sticks.

Why do most AI projects in L&D stall?

Because they start as tool trials rather than capability decisions. Somebody buys licences, runs a pilot, produces impressive artefacts, and nothing changes in the operating model.

The pilot ends. The old process reasserts itself within weeks, since nobody changed who reviews what, or what quality bar the output has to clear. I covered that failure pattern in why AI in L&D is failing most companies.

However, the quieter reason is trust. Teams will not put a generated draft in front of a client until somebody defines what good looks like and who is accountable when it is wrong.

How do you judge an AI L&D consultant?

Ask what they would tell you not to automate. Anyone who cannot name three things has not thought about risk, accuracy or the parts of design that carry your reputation.

Then ask how the capability transfers. Specifically, at the end of the engagement, can your own designers run the workflow without the consultant in the room? If the answer is vague, you have hired a dependency.

The tell is who owns the prompt library, the standards and the review gate at the end. If that sits with the supplier, you rented a capability instead of building one.

What does readiness actually mean?

Readiness is mostly unglamorous. Clean content you can point a model at, a documented quality standard, a review process with named owners, and clarity on what your data can legally touch.

Notably, none of that requires a single licence. I set out the detail in critical L&D AI readiness assessment steps, and the order matters more than the tooling.

Get those four in place and almost any competent stack works. Skip them and the best stack in the market will still produce inconsistent output nobody trusts.

Should an AI L&D consultant build your agents?

Sometimes, and later than you think. Agents are genuinely useful once the task is stable, the standard is explicit and somebody owns the output. I argued that case in AI agents in L&D are already your learning platform.

Before that, an agent just automates an argument you have not settled. As a result you scale the inconsistency rather than the work.

In addition, watch the maintenance question. Every agent is a small product with an owner, a version and a failure mode. Two of those turn up in the budget. All three turn up in reality.

How should this be priced?

Short, sharp and outcome-led. A diagnosis engagement should take days, not months, and it should end with a decision you can act on rather than a sixty-page report.

If a proposal runs for two quarters before anything changes in your workflow, that is a retainer wearing a strategy costume. Ultimately you are paying for judgement, and judgement gets delivered quickly.

The right AI L&D consultant leaves you with fewer tools than you expected, a clearer standard than you had, and a team that can run the thing themselves.

Everything else is theatre with a good demo attached.

If you want that diagnosis first, book a Discovery call at calebfoster.ai and bring your real bottleneck rather than your tool shortlist.