Key takeaways
- There is a self-preservation economy inside advisory work - some advisors push to keep a client off AI because a client who can ask its own systems questions no longer needs a gatekeeper.
- Watch whether the caution tracks the billing model - when the warning gets loudest exactly where it would shrink the hours, the warning is about the invoice, not the risk.
- Why does this happen at all? Hourly billing creates a plain disincentive to make a client faster, because faster means fewer hours.
- The straight version of the work gets more useful with AI - the interview that surfaces what you don't know you don't know, pricing on outcomes, teaching you to run it yourself.
An outside advisor once fought, hard, to stop a company from plugging AI straight into its own systems. The reasons sounded careful. Too risky. Not ready. Data isn’t clean enough. Here’s the part that gave it away: the resistance tracked the advisor’s billing model far more closely than it tracked the company’s actual interest. That’s the whole post in two sentences. When a client can sit on top of its own system of record and ask it questions in plain English, the system stops being a black box that only the advisor can open. It becomes, in blunt terms, a glorified database with accounting rules. And anyone can query that.
So the title answer, up front: the consultant fought AI because AI removes the lock he was holding the key to. Not every advisor does this. Plenty are worth every pound. But the incentive is real and quiet, and once you’ve seen it you can’t unsee it.
I’m going to aim the blade at the incentive, not the profession. Big difference.
Where the resistance really comes from
Economists have a clean name for this. The principal-agent problem. You hire someone to act for you, but they’ve got their own incentives, and those incentives don’t always point the same way as yours. The textbook version puts it plainly: the principal-agent problem is a situation where an agent is expected to act in the best interest of a principal, but the agent has different incentives, leading to a conflict of interests. The engine underneath it is asymmetric information. The agent knows more than you do.
In advisory work the asymmetry is the product. You pay for the gap between what the advisor knows about your messy data, your half-documented processes, your weird integrations, and what you know. Close that gap and the meter slows down.
Now drop an AI layer on top of the system of record. Suddenly a finance lead can type “show me every invoice over a threshold that skipped approval last quarter” and get an answer in seconds. No advisor required. No three-week engagement to “scope the reporting requirements”. The asymmetry collapses. And the person whose income depended on that asymmetry feels it in their gut before they can explain it in their head. I think that’s what’s actually happening in a lot of these “too risky” conversations. The objection is sincere. It’s just downstream of a threat the advisor hasn’t named to themselves.
Mind you, I’ve been wrong about people’s motives before. So I try to test it rather than assume it.
Why would a good advisor fight your independence?
Because the billing model rewards your dependence. This isn’t a character flaw, it’s arithmetic. When you charge by the hour, you make less money by making the client faster. Greg Lambert put it bluntly on his law blog in October 2025: when a firm charges by the hour, there is a disincentive to reduce hours spent. Read that twice. The structure punishes the exact efficiency the client is paying to get.
The legal world is living this out loud right now because it bills the same way. Everlaw surveyed 299 legal professionals and the numbers aren’t subtle. Lawyers using generative AI report reclaiming up to 260 hours a year, which is 32.5 full working days per person. And 90% of them think AI has already changed how legal work gets billed, or will within two years. So the time savings are real and large. The catch? Saved hours only shrink the bill if the firm chooses to pass them on. A lot of the savings just get quietly reallocated to other billable tasks. The hours move. The invoice doesn’t.
Apply that to AI advisory. If the advisor’s model is “I bill you while you depend on me to interpret your own systems”, then teaching you to query those systems yourself is, from a pure revenue angle, an act of self-harm. Of course some of them resist. The surprise would be if none of them did.
Does that make every cautious advisor a rent-seeker? No. Next question.
Spot the self-preservation tell
Here’s the practical bit, the thing you can actually use on Monday. You can’t read an advisor’s mind. But you can watch where the caution clusters. Real risk caution is roughly evenly spread. Self-preservation caution is suspiciously specific. It gets loud precisely at the points where your independence would cost the advisor hours, and goes quiet everywhere else.
Run this short test the next time someone tells you a direct AI integration is “not ready”:
- Ask what would make it ready, in writing. A real risk has a checklist and an exit. A protected revenue stream has a moving goalpost that never quite arrives. If the answer keeps drifting, that tells you something.
Watch for three more patterns. The advisor who insists every query must route through them “for governance” but can’t explain a governance rule a tool couldn’t enforce. The one who treats your own data as their proprietary knowledge. The one whose estimate for “let your team self-serve” is mysteriously larger than the estimate for “keep paying us to pull reports”. None of these is proof on its own. Together, they rhyme. When consulting with companies on this, the cleanest tell is simple: ask who owns the queries at the end of the engagement. If the answer isn’t “you”, ask why not.
What the straight version of the job rewards
Now the fair half, because the cynicism only goes so far and I don’t actually believe advisors are the villains here. The clean version of advisory work gets more useful with AI. Quite a lot more.
Think about what a good advisor really does. The best ones run the interview that surfaces what you don’t know you don’t know. That skill doesn’t get automated, it gets amplified, because once the boring data-pulling is gone you’ve got more room for the hard human questions. They price on outcomes, not on time served, so efficiency becomes their friend instead of their enemy. They teach you to fish. The whole point of outcome-based engagement structures is that the advisor wins when you win, which flips the incentive the right way round. That’s the companion piece to this one. This post is about the psychology, that one is about how to wire the contract so the psychology stops mattering.
There’s also a strong argument that AI won’t kill good advisors at all. Dave Friedman made the case in a piece on why AI will not kill McKinsey: the elite firms sell context and judgment more than raw analysis, and even as the analysis gets democratized, the decision rights won’t be. Boards still want a human to bless a hard move and absorb the blame if it goes wrong. AI changes the inputs. It doesn’t change who carries the political risk of a layoff or a merger. I think he’s mostly right. The advisor who sells judgment, courage, and accountability is safe. The advisor who sells access to your own information is the one sweating.
Funnily enough, this is the same pattern I keep seeing on the product side. In building Tallyfy, the teams that thrive with automation are the ones who treat the tool as something their own people operate, not something a vendor holds hostage. The ones who outsource understanding stay dependent forever. AI just makes that dependency optional in a way it never was before, which is exactly why the gatekeepers are nervous.
Be fair to the advisor who teaches you to fish
Picture two advisors walking out of the same kickoff meeting. One has spent the hour mapping which of your questions a tool could answer without them, so they can charge you for the questions only a human can. The other has spent the hour building reasons you’ll always need to call first. Same suit, opposite business model. Your job is to tell them apart, and the AI layer on your own systems is the cheapest test you’ll ever run, because it reveals who’s afraid of you knowing things.
This connects to a broader pattern in why AI projects fail: the failure is rarely the model. It’s the incentives and the process around it, and an advisor quietly steering you away from owning your own systems is a process failure with a friendly face. If you’re setting up an advisory practice of your own, by the way, the lesson runs the other direction too. Build the kind that gets stronger when clients get smarter. The rent-seeking kind has an expiry date now, and the date is getting closer.
So keep the advisor who hands you the rod. Pay them well, because the interview that finds your blind spots is worth real money and always will be. Drop the one guarding the key to a lock that no longer needs to exist. The market is about to make that choice for everyone anyway. Might as well make it on purpose.
