The Veneer of Competence

How AI is making mediocre work look brilliant and what to do about it

You will have seen this report; sharp subheadings, great formatting, perfect length… a few too many em-dashes and commas everywhere but its executive summary is confident and at a glance it looks amazing.

Then you actually read it.

The argument goes in circles, key facts are missing or just plain wrong, key claims contradict figures, the list goes on.  The recommendation sounds decisive but It’s cobbled together nonsense, stitched from plausible-sounding fragments.

I have started calling this the veneer of competence: work that looks superficially excellent but falls apart the moment you scratch the surface. Like a thin layer of quality wood glued over cheap chipboard, it’s built to be looked at, not sat on.

How It Happens

The veneer isn’t new. Since cavemen were trying to get out of fighting tigers by writing up reports on berry collection, people have dressed up thin work with confident language and sleek format.  AI has just changed the cost of this piece of theatre. Polish used to take effort by the author and they may even have had to understand the material but no longer.  Anyone can now with a few prompts pull together a report on anything and at a glance it looks plausible.

An employee who doesn’t know the subject uses their company AI-assistant to pull together a fluent looking and self-assured report. There is no hedging or stating that there are guesses, it is full on ‘I don’t need a map’ energy.

They skim read it back and it looks fine.  They may even go a step further and ask a second (or the same) AI tool to check the report, which sounds like diligence but isn’t exactly an independent verification.

And here’s the cruel part: the less you know about a topic, the harder it is to spot what’s wrong with a convincing answer about it. The people most likely to over-rely on AI are often the ones least equipped to catch its mistakes, Dunning Kruger on steroids.

Why Does It Occur?

It is tempting to cast the person who produced the report as the villain and sometimes they are. They talk a good game but don’t actually know much about it but are good at the management lingo. Other times they are simply responding to what they see rewarded.

They also know what happens next because they have seen it before. The report gets a glance from management, ‘great work’, then they can dine off this for a week while sloping off for coffee and doing little else and there is little management scrutiny, no testing of the polish. So they learn, as does everyone else, that polish is the only thing. Style over substance.

With AI adding to this, the shiny report takes minutes, more time for coffee. A properly researched and checked one would take days and the extra effort earns nothing the glance can see. Even a conscientious employee learns quickly that the week spent verifying sources buys no extra credit and can often land them in trouble for taking longer than their colleagues.

They are not gaming the system so much as responding to it: management want a shiny report, management don’t read it and management don’t question it, so a shiny report is exactly what management get.

I’m not excusing the behaviour but it does explain how it’s come about and will stay until the rewards change.

The Glance Test

The report then goes to leadership, who are busy. They glance at it and it looks amazing. Then they say so, loudly, in front of the whole team.

Meanwhile, the colleague who actually knows the subject opens the same document and starts tutting after the second paragraph. They see glaring errors, the invented nuance, the muddled logic, the confident statements with nothing underneath. But they’re not the audience that matters and they know it.

That’s the real problem: the people who can see the flaws are rarely the people who evaluate the work. Managers are short on time and rewarded for throughput. A document that looks right gets treated as right.

So, What Are The Consequences?

This is the question that bothers me most, because the honest answer is: often, none. At least not for the person who produced it.

The consequences land elsewhere:

• Promotion Without Substance. The person with the glossy output gets noticed, praised and promoted ahead of colleagues with genuine depth, who quietly do the harder thinking and produce less flashy work.

• Resentment. Everyone knows but nobody says it, apart from at lunch, where the rest of the team are furious. The kiss-up, kick-down office is alive and well, while knowledgeable team members watch the reward of appearance over ability and quietly disengage.

• Bad Decisions. Somewhere down the line, someone acts on that cobbled-together analysis. The cost arrives later and is rarely traced back to its source.

• Eroded Trust. Sometimes the hand burnt learns that that polished documents can’t be trusted, everything has to be re-checked and the efficiency AI promised quietly disappears. 

What Can You Do About It?

While venting with your sympathetic colleagues at the proverbial water cooler is cathartic, it doesn’t actually solve anything.  It makes you look bitter and the culture that rewarded the veneer will still be there on Monday.

The more productive move is to make sure you’re never the one producing it. That means using AI deliberately rather than abdicating your critical thinking. The aim isn’t to use AI less, it’s to use it well. A useful framework for using it well is the “4 Ds” of AI fluency, developed by Rick Dakan and Joseph Feller: Delegation, Description, Discernment and Diligence.

1. Delegation

Decide what to hand over and what to keep. AI is excellent at drafting, summarising, reformatting and generating options. It’s much worse at judgement, context and knowing what matters to your organisation. If you’re delegating the thinking itself, you’re not delegating, you’re abdicating.

2. Description

Rubbish in, polished rubbish out. This aligns with previous points I have made about having good data. Vague prompts without detailed context and guardrails produce generic waffle. It fills the page but has no substance. Give the prompt details, tell the tool who the audience is, what the goal is, what constraints apply, what you already know and what good looks like. The better direction and description you provide, the more likely it is that you understand it, which is rather the point.

3. Discernment

This is the one that separates a veneer from the real thing. Read the output as a sceptic, don’t just skim read. Does the argument actually hold together? Is every claim something you could defend if challenged? Could you explain it without looking at the page? Asking a different AI to check can be a helpful option but so is asking a human who understands the subject to read it through and give you their feedback.

4. Diligence

Take responsibility for what you put your name on. Verify facts and figures against real sources. AI can cite sources that are out of date or from less reputable places – Reddit is great but I wouldn’t want to use it as a source for statistics. Own the result, your name is on the document and regardless of whether you typed it or not you are the human behind it. Would you sign a tax return without checking it first?

A Note For Leadership: This One Is On You

We know you’re busy. Reports are arriving faster than anyone can find room in their diary to read and discuss them. So ask yourself: why did you request a report you don’t have time to read? Is it so you can say it’s been done? An audit tick-box, with no time to look at what it implies?

You are the ones that determine the culture of your organisation.  When a polished document is praised in the team meeting after a cursory skim-read, everyone learns what gets rewarded. Culture and trust in leadership is eroded slowly over time.

The colleague who knows the subject notices and so does the person who spent a week getting the thinking right. Repeat this often enough and that’s the culture you will get - style over substance and your best people will disengage or leave.

How to stop the rot

• Read Before You Praise. Hold the applause until you’ve tested the work.

• Ask The 4 D Questions. Ask for the reasoning not just the result. What did you challenge? Someone who used AI well will answer easily but someone who didn’t will struggle within a couple of minutes.

• Make Peer Or Second-reader Review Normal. Normalise reviewing and commenting on others’ work, so no one feels like they’re being singled out. A culture of collaboration will guard against reports with glaring issues.

• Reward Visible Thinking. Moving to a culture of psychological safety, where it is safe to have honest opinions, state sources, caveats and flag uncertainty.

• Set The Standard Out Loud. Using AI is welcome and using it well is the expectation. Share the 4 Ds with your team and make that the way work gets done, including being open about where AI helped.

• Act When You Find A Veneer. Raise it early and privately, as coaching first. If it repeats, treat it as a performance issue. Ignoring it tells the whole team the standard doesn’t exist.

Do all of this and you are starting to move into the thermocline of truth territory (I will probably write a blog about this too).

AI has made polish free. Your attention is now the bottleneck that tells good work from good-looking work.

The Bottom Line

AI is a remarkable and powerful tool. Used well it makes good people better but it has made the appearance of competence nearly free, which means appearance can no longer be our measure of competence. The veneer will keep fooling people for as long as we reward the glance rather than the read.

So make sure you’re aware and not being fooled by cheap veneer.

Reference: Dakan, R. and Feller, J. (2025) The AI Fluency Framework. Copyright Rick Dakan, Joseph Feller and Anthropic, released under CC BY-NC-SA 4.0.