Somebody inside an organisation decides this work is worth doing, and then has to defend it to a finance director, a chief executive or a board. They are not in the room with me when that happens. This page is written for them rather than for me, which means each objection is given its strongest form and the answer says where it is right.
Three of the six are good objections. One of them should stop a purchase outright.
This page describes a category of work I am paid for, so the conflict is obvious. The test to apply to all of it: every answer below should be checkable against something other than my saying so, and where it is not, the page says that too.
One. Why not just buy the tools and let people work it out?#
Often you should, and this is the objection that should stop a purchase. If a team needs to draft faster and the work is checked by somebody who could have produced it, buy the licences and leave it alone. Most AI use in most organisations is exactly this, it is fine, and paying anybody to think about it is waste.
The objection fails on a narrow class of work: decisions where the person checking the output could not produce it themselves. That is the test, and no other one matters here. A plausible wrong answer and a correct one look identical to a reviewer who cannot independently evaluate the content, so the review step is adding a signature rather than a check.
Vaccaro and colleagues, in a meta-analysis of 370 effect sizes across 106 experiments, found human and AI combinations performing worse on average than the stronger of the two alone, with the losses concentrated in decision tasks. That finding is about controlled experiments rather than your organisation, and the averages hide large variation. It is enough to say the pairing is not automatically additive, which is what "let people work it out" assumes.
So: buy the tools for production. Think about it where the output is a decision and the checker is weaker than the system.
Two. Why not just run AI literacy training?#
Partly right, and literacy is a legal requirement rather than an option. Article 4 of the EU AI Act has required a sufficient level of AI literacy since February 2025 across every risk tier. If you have no programme you have a compliance problem before you have a capability one, and a provider who can run one is the correct purchase.
Where it falls short is what gets measured at the end. A literacy programme reports completion. Completion says somebody attended. It does not say anyone can now do the work unaided, which is the thing that was supposedly at risk. Those are different measurements and almost nobody takes the second one.
The sharper version of the objection is that capability work is literacy training with a more expensive name. The answer is a testable difference: a literacy programme ends with a completion rate, and capability work ends with a number for what people can do without the system, gathered the same way twice. If somebody sells you capability work and reports a completion rate, they sold you literacy training.
Three. We already have an AI policy and a strategy.#
This is the weakest of the six and the most commonly offered. Almost every organisation has both. The question a policy does not answer is which specific decisions a machine may now make, who can stop each one, and how anybody would find out if it had gone wrong.
A policy is a statement about conduct. The thing missing is a statement about particular decisions, which is the only form a board can hold anyone to. The gap is visible in about ninety minutes with the delegation boundary map, which is published in full and free to run without me.
If you run it and every stage already has a named owner, you do not need this work. That is a real possible outcome and it has happened.
Four. Our people use it every day, so we are fine.#
Usage is the measure most likely to be wrong in the reassuring direction. Seat counts, licence take-up, prompt volumes and adoption percentages all measure activity. None of them says anyone got better at anything, and all of them rise when an organisation is drifting as fast as when it is designing. The pattern has a name on this site: usage theatre.
Where the objection is right: high usage does tell you the tools are not being rejected, which is a real and common failure mode and worth knowing you have avoided.
The follow-on question that separates the two: how many times has anyone overridden the system this quarter? Zero is the worrying answer, because it evidences an untested right rather than a good system.
Five. Why not use the provider we already have?#
Usually you should. An incumbent who knows your business, your systems and your people starts with advantages no outsider can buy, and the switching cost is real. If they can do it, use them.
Two questions to put to them before you do. What would you tell us not to do, and where does your revenue come from if we follow your advice? A firm whose income comes from building has a commercial interest in the answer being a build. That is the shape of the business rather than dishonesty. The comparison between an independent advisor and a consultancy turns on incentives rather than on ability for that reason.
My own answer to the second question, since it applies equally: I am paid for advice, so I have an interest in you concluding that you need advice. What I have no interest in is which tool you buy or how large the programme becomes.
Six. Is this not the usual human-skills material with AI bolted on?#
A fair suspicion, and the market earns it. Lists of human skills have been sold for thirty years and most of them are a taxonomy with a workshop attached.
The difference is testable rather than rhetorical. A skills list is a statement of demand, usually gathered by asking employers what they value: the World Economic Forum's series is the most-cited example and it measures no worker and tests no skill. A capability claim is a statement about what a person can do when the system is taken away, and it can be checked. One tells you what to put in a budget. The other tells you what to test for, and it carries the harder obligation. The full comparison is at SuperSkills and the WEF Future of Jobs Report.
Where the suspicion lands: I have not published a validated instrument for the seven, and nor has anybody else for their list. Until somebody does, both are arguments, and mine has less survey data behind it than the Forum's.
What would settle it#
A controlled comparison of capability work against literacy training against doing nothing, in matched organisations, measuring unaided performance at two years rather than satisfaction at the end of the session. Nobody has run it, in this field or near it. Anybody quoting you a figure for the return on this category of work is giving you a sales claim, including me.
How to use this page#
Take it to the meeting rather than to me. If objections one or five hold in your case, you should not buy this work, and I would rather you reached that conclusion from a page than from an invoice.
- The one question worth more than the rest: could the person checking this output have produced it themselves? If yes, across the decisions that matter, stop here.
- The cheapest test: run the delegation boundary map on one real decision. It is free, it takes ninety minutes, and the gaps are the finding.
- The thing to ask any provider, me included: what would you tell us not to do.
Where this sits in my own argument#
The rest of this research argues that judgement rather than production is the constraint. An objection page is that argument turned around and pointed at the purchase, because an argument worth making should survive somebody trying to spend nothing on it. Three of the six above are good enough that an organisation acting on them would be right, and a page that pretended otherwise would be the thing it is warning about.
Related SuperSkills research#
On the measure that flatters, usage theatre. On what a policy leaves undecided, design versus drift and the AI readiness lie. On testing rather than counting, how to assess capability rather than output. On the sequence, rules before tools. On who to hire, an independent advisor or a consultancy.
Key sources
- Vaccaro, M., Almaatouq, A. and Malone, T. (2024). When combinations of humans and AI are useful.
- Dell'Acqua, F. et al. (2023). Field experimental evidence on the jagged technological frontier.
- World Economic Forum (2025). The Future of Jobs Report 2025.
About this research#
Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), independent advisor to chief executives and boards, and founder of The SuperSkills Intelligence Company.
This page argues for and against a category of service I sell. The conflict is stated in the opening rather than disclosed at the foot.
How this research works · Reviewed quarterly · Found an error? Tell me and it is corrected on the page.
Essay · SS-2026-401 · 1 peer-reviewed study, 1 working paper and 1 institutional survey
Hirji, R. (2026). Six Objections to This Work. The SuperSkills evidence base, SS-2026-401. https://thesuperskills.com/research/six-objections-to-this-work. Last reviewed 3 October 2026.
An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.
How citations and IDs work