The Signal #019 — Dakota’s read on the AI news that actually matters to people running a business.

A pattern is spreading through boardrooms and all-hands meetings right now. The CEO gets access to an AI tool, builds something that works, and immediately starts doing math on headcount. It feels logical from where they sit. It is not.

What happened

Mike Masnick at Techdirt published a piece on June 9, 2026, describing something he has now seen four separate times in the last three months. Each case follows the same script: a CEO sends an all-hands email declaring that LLM tools (large language models, the AI systems behind tools like ChatGPT and Claude) are mandatory, that everyone must start using them immediately or find work elsewhere, and that consultants, office hours, or internal hackathons are being set up to force adoption.

The worst version of this, according to Masnick, is companies that set up token leaderboards. A token is a small chunk of words the AI reads or writes. Tracking who uses the most tokens as a sign of good AI adoption gets the incentive exactly backwards. Good AI usage means treating tokens as a scarce resource, not burning through them to look productive on a dashboard.

Box CEO Aaron Levie, described in the piece as a genuine AI believer, put a sharper point on why this keeps happening. His argument: CEOs are uniquely prone to what he calls AI psychosis because they are sufficiently distant from the last mile of work that still has to happen to generate most value with AI. When a CEO plays with an agentic coding tool (software that can take a goal and execute multi-step tasks to reach it), they see the happy path. They do not see the next ten or twenty things that have to happen before the output is actually usable. They do not review the code before it goes to production. They do not verify the contract terms before it goes to a counterparty. They do not deal with security review, legal compliance, or accessibility requirements. They see a working prototype and conclude the people who handle everything else are now redundant.

Mansnick’s framing is worth keeping: making things work is different than making things work well, at scale, in a specific environment, for a mass market.

Why it matters for operators

If you run a mid-size e-commerce operation, a marketing agency, or a professional services firm, you have probably felt some version of this pressure. Maybe it came from a board member who saw a demo. Maybe it came from a competitor who announced layoffs framed as an AI efficiency play. Maybe it came from inside your own head after you watched an AI tool draft a solid first pass on something that used to take your team hours.

That pressure is not entirely wrong. These tools do change what individuals can accomplish. A solo account manager at a SaaS company using AI well might genuinely cover ground that used to require two people. That is real.

But the leap from that observation to firing half the staff skips over everything your experienced people are quietly doing that never shows up in a demo. The analyst who catches the data anomaly before it becomes a client complaint. The ops manager who knows which vendor will actually deliver and which one will not. The support lead who can read between the lines of a customer ticket and escalate before it turns into a churn risk. None of that is visible in a clean AI prototype. All of it still matters.

What most people get wrong

The Techdirt piece makes a point that is easy to miss. Masnick writes that no one who is forced into using these tools will ever learn to use them well. Mandatory AI adoption, backed by threats, produces compliance theater. People will use the tools in whatever way satisfies the metric, not in whatever way actually improves the work.

Token leaderboards are the clearest example, but the problem is broader. Any operator who builds their AI rollout around surveillance and pressure rather than curiosity and real workflow improvement is going to get performative usage and resentful employees. They will not get the productivity gains they were sold on.

The better framing, also from the piece, is that the best case for these tools is building personalized tools to assist you in accomplishing a specific task, not building mass market tools or replacing the people who handle the details that matter at scale. That is a much quieter, slower, and more durable kind of adoption than a big all-hands email can create.

The closing lesson

AI tools are worth understanding and worth adopting thoughtfully. The operators who will actually get value from them are not the ones issuing ultimatums. They are the ones asking their teams where the friction is, then testing whether AI can reduce it. That takes longer. It also works.

The CEO who sees a working prototype and concludes the team is optional has not discovered an efficiency. They have just revealed how little they understood the job in the first place.

If you are thinking through what AI adoption actually looks like inside your operation, without the hype and without the pressure tactics, the resources at xovionlabs.com are built for that kind of practical thinking.