The gate was there for a reason

(This is part two of a two-part piece.)

In part one, I used the bixonimania story to make one point - removing a constraint that could move, without asking why it existed, is where things go wrong. That gap - between 'can' be changed and 'should' be changed - is what this piece is about.

Let's use AI to look more closely at the gap, because it's the place right now where most people are making this mistake without noticing.

## Level one: your personal life

This is the lowest-stakes layer, and also the one where AI has earned its place with the least argument.

People are building things here that would have required hiring someone or knowing how to code two years ago.

- A birdwatcher building a personal sighting tracker with photo logging.

- Someone flat-hunting in a new city who's built a tool that filters listings against a set of criteria and flags anomalies.

- A person who reads obsessively and has built themselves a book recommendation engine trained on their own history.

- An AI that manages their calendar, drafts responses to routine messages, reminds them what they said they'd do and didn't.

These are real uses. They're genuinely good. The constraints here are mostly yours to set, and most of them are negotiable. If the book recommendation is wrong, you read something mediocre. The feedback loop is fast, the cost is low.

The one constraint that isn't negotiable, even here - your judgment stays in the loop on anything that touches another person. Not because AI is always wrong. Because the cost of being wrong there is not recoverable the way a bad book recommendation is.

That gate should stay human. The question when using AI at this level isn't whether to use it or not, it's whether you've named which decisions remain yours.

## Level two: your individual professional life

This is where most people are under-thinking it, because they haven't mapped which gates matter.

Take a salesperson. The place where AI is genuinely useful is in absorbing the administrative weight that has always been the tax on doing the actual job. The CRM updates that eat up thirty minutes regularly. The prep sheet before a meeting - pulling together what's publicly known about the company, recent news, the last three interactions, open questions from the previous conversation. The follow-up email structure after a complex discussion. The win-loss summary at the end of a quarter. The agenda for a QBR. This is real time returned to the person, and it matters.

What AI does not do well - and where people are currently burning goodwill and pipeline - is generic outreach at scale. The assumption that a first draft personalised with a name and a company and a LinkedIn post reference will cut through is wrong. And it's getting more wrong every month, because everyone has the same tool producing the same draft. Your buyer can feel it. The constraint that cold outreach only works when it demonstrates you actually paid attention - that one hasn't moved. You're just pretending that you can go around it.

Similarly, account research sounds like an obvious AI win until you try it without a well-built template and a clear brief. Raw AI research gives you what's publicly available and generally relevant. What a good rep actually needs before a strategic conversation is specific, opinionated, and built on pattern recognition from dozens of previous conversations in that sector. That doesn't come from a prompt. It comes from doing the work enough times that you know what to look for. AI can accelerate retrieval. It can't replace the judgment about knowing what to look for.

The non-negotiable gate at this level - human judgment on a relationship.

What a specific person actually cares about. Whether the moment is right. Whether the tone of that message is going to land or cost you the deal. No model knows your buyer. You do.

That gate is yours to manage. Some gates are not.

I'm going to use a pharma example here because it makes the asymmetry vivid. A marketer for a tech company who publishes something suboptimal can take it down. Gates don't have to be human-managed. A marketing team at a CDMO handling artwork before print cannot just roll it back so easily. Every word on that label carries legislative weight. An error doesn't mean a recoverable mistake - it means a product recall, regulatory action, potentially patient harm. You can use AI to accelerate plenty of things here. The gate, the sign-off must be human-owned. That is not negotiable there, not even slightly. The efficiency gain from relaxing it is irrelevant against what breaks if it fails.

The question when using AI - have you named which gates stay human, and why?

## Level three: leading a team

This is the level where the most damage gets done, and also where the most leverage exists if you get it right.

A VP of Sales deploying AI across a team isn't just changing what individuals do. They're redesigning the workflow - what gets checked, what gets skipped, what becomes habit, what disappears without anyone deciding to remove it.

Take the pipeline review. Every VP of Sales runs one. The AI-assisted version can now pull together a pipeline summary faster than any ops person could - deal stages, velocity, coverage ratio, forecast versus target. That's useful. But the question the AI cannot answer is - which of these deals is real? Not real in the CRM sense - real in the sense of whether the champion actually has internal support, whether the timeline is genuine or aspirational, whether the competitor you've been told isn't in the deal actually is. That read comes from the rep's lived experience of the account and the VP's pattern recognition across a hundred previous deals. The moment you let the AI-generated pipeline summary substitute for that conversation, you've removed a gate that was load-bearing.

Win-loss analysis has the same shape.

AI can synthesise call transcripts, CRM notes, and exit interviews into a clean summary of why deals were won or lost. That's valuable - it uncovers patterns faster than any manual analysis. The constraint that cannot move - a human has to interrogate those patterns against context the system doesn't have. The deal you lost because of price might actually have been lost because your champion is in the process of exiting. The AI will tell you it was price because that was the official reason given. The VP who was paying attention knows the real story.

The question for anyone leading a team right now is not "where can AI make us faster?"

The better question is, if we move fast here, which gates are we implicitly removing, and have we consciously decided those gates should go?

Most organisations haven't had that conversation. They've adopted tools, watched certain things get faster, and haven't noticed what stopped being checked.

The researcher who ran the bixonimania experiment was careful about her conclusion - AI didn't create the failure. The failure was human. People stopped being critical about sources because a tool made it easy to stop.

Which of your gates are actually non-negotiable? Do you still know where they are? And when did you last check?

Part two of the bixonimania piece. The framework it applies is under How We Work.

Read part one here.