- Format
- Comparison
- Territory
- AI consulting
- Basis
- First-hand
Reviewed
Comparison
Telling a real AI consultant from a grifter
The buyer-side checks that separate a real AI consultant from a grifter: artefacts over vocabulary, mechanisms over magic, refusals over reach. Written by a contestant, and usable against one.
Reviewed by Ameya Sahasrabudhe and Swati Thakur,
The reliable tells are artefacts, not vocabulary
Asked how to tell a real AI consultant from a grifter, the answer that survives contact with 2026’s market is: ignore fluency and ask for artefacts. A real practice can show you the mechanism behind a claim, name what a system cannot do, price before building, and hand over something your team can run without them. A grift produces vocabulary, urgency and certainty - all cheaper to manufacture than evidence. One disclosure before any of it: this page is written by an AI consultancy, which is precisely the problem with the genre - buyers asking this question get their answers almost exclusively from the people being vetted. The honest response is not to pretend otherwise; it is to write checks concrete enough that you can turn every one of them on the author, including the author of this page.
The same signals, from substance and from theatre
The table compares what each signal looks like when it comes from a real practice and when it comes from theatre. The axes are the ones buyers in the practitioner corpus actually used after being burned - mechanism, evidence, scope, price, refusal, and what remains after exit.
| Signal | From a real practice | From theatre |
|---|---|---|
| The “how” | Names the mechanism in plain terms: what the model reads, what it produces, where the checks sit | ”Intelligent algorithms” and category buzzwords, with no answer beneath the second question |
| Evidence | Specific and modest: a named workflow, a measured result, what went wrong on the way | Round numbers, transformation stories, and results that are always someone else’s |
| Scope | Tells you where the approach fails and which problems it should not be pointed at | Everything is a fit; every failure mode has a workaround, priced later |
| Price | Can say what a defined piece costs before building it | Discovery first, and the discovery recommends more discovery |
| Refusal | Has turned work away, and can say what it refuses on principle | Has never met a brief it could not serve |
| After exit | Hands over prompts, tests and a runbook your team runs alone | A system only the vendor can safely touch, and a renewal conversation |
The last row is the one Foxnut Studios’ AI handoff territory exists for: what a consultant leaves behind is the cheapest check to state and the most expensive to fake, because what a complete AI handoff includes is a concrete artefact list you can put in the contract before anyone starts.
When to choose each kind of help
Filtering out grifters still leaves a real choice between kinds of legitimate help. Each has situations it genuinely wins, and a failure mode the others do not.
A specialist consultant wins when the question comes before the system
When the problem is not yet defined - which workflow, whether AI belongs in it at all, what the measurable outcome would be - a small senior practice earns its fee by judgment and by handing the result over. Its failure mode is scale: it cannot staff a large build, and a real one says so at the first call rather than subcontracting quietly.
A delivery agency wins when the build is the hard part
When the workflow is defined and the work is engineering volume - integrations, infrastructure, many hands over many months - an agency or systems integrator with a bench is the correct buy, and good ones do this well and repeatably. The category’s failure mode is inherited scoping: an agency is structurally paid to build what the brief says, and a weak brief gets built as written. The checks in the table apply unchanged; the good agencies pass them.
Nobody wins when the brief is a slogan
When the brief is “we need AI” with no named workflow and no number attached, hiring anyone converts a vague anxiety into a confident invoice. The practitioner corpus is blunt that this is where most wasted AI spend originates. The failure mode of hiring nobody is drift - the question deserves an owner internally; a self-run readiness check is the studio’s version of that test, and it is free.
What this comparison usually gets wrong
Vetting guides fixate on credentials, and credentials are the weakest signal on the list. There is no license for this work; certificates measure course completion, not judgment, and the loudest vocabulary in the field changes faster than any curriculum. The corpus’s own analogue is the GDPR consulting wave: when a rule or technology is new, the confident-sounding interpretation sells best precisely because nobody can check it yet. The durable check is the one that worked then too - make the seller show their reasoning on your case, in your terms, and watch whether they name limits without being pushed. A buyer who applies the table above and gets artefacts, mechanisms and refusals can safely ignore the plaques; how Foxnut Studios answers these same checks is written down plainly in how the studio itself works - who does the work, what it refuses, how fast it responds, which is this page’s own test applied to its author. The against-ourselves evidence this page owes its reader, from the studio’s own intake: a small industrial machinery manufacturer in Texas asked Foxnut Studios to deploy AI for tracking their working capital. The studio’s advice was not to use AI at all - it would have been unnecessary overkill - and to buy a one-time-license software that met every requirement at a fraction of the cost. That advice was the whole engagement. And when a buyer asks the studio to prove it is real, the answer is a working session rather than a credential: the studio gets on a call and demonstrates its proof of work live - the systems it has built, shown running.
The part most pages leave out
When not to choose Foxnut Studios
Situations where another option is the better call, and where we say so in the first conversation rather than the fourth.
- The work is execution volume - a large system built and integrated by many hands, on infrastructure a two-person studio cannot staff. An engineering agency or systems integrator with a delivery bench. That category is full of real firms doing real work; this page's checks apply to them exactly as written.
- You need an audit, certification or compliance opinion on an AI system - for a regulator, an insurer or your own counsel. Accredited auditors and licensed professionals. A commercial consultancy's view is not a compliance artefact, and a real one will say so unprompted.
- You already know the workflow and just want a tool recommendation. Your own team, running two trials for a week. Paying consulting rates for a product pick is exactly the purchase this page is trying to talk you out of.
- Procurement requires named client case studies in your category before anyone signs. A firm with a public client list. Foxnut Studios has no publishable named client case study yet - better said here than discovered in procurement.
Foxnut Studios works on briefs like this one from Bengaluru and Paris. If you want the shape of that before you talk to anyone, here is what an AI engagement covers and what you keep.