Antibody Response

In the spring of 2026, a Google engineer built a tool in what looked like a handful of spare weekends. It did something simple. It let you run Google Workspace — Gmail, Docs, Drive, Calendar, the whole suite — from the command line: a plain text box where you type one instruction instead of clicking through menus. Engineers call it a CLI. And because it was just text in and text out, an AI agent could operate it too — no menus, no clicking, just an instruction and a permission. Developers loved it. It climbed to the top of Hacker News — the tech industry's front page — and was written up, straight, by the technical press.[4] The code now lives under Google's own GitHub organization, with tens of thousands of stars and a one-line disclaimer that it is “not an officially supported Google product.”[1] The engineer who built it, by his own account, was fired for it.[2]

The threat was never the tool. It was what the tool made impossible to un-see.

Here is the part you can check without trusting anyone's motives. Days earlier, at Cloud Next — its largest developer conference — Google had announced from the main stage that an official Workspace CLI was coming, one built to let you “interact with these capabilities directly from your agents.”[3] So the sequence, reduced to what is verifiable: an employee shipped the agent-native version of the product first, it became the most-loved thing in its category for a week, the company announced its own — and the employee was gone. The motive that ties those together is his word against a silence; Google never commented. But the shape is one of the most reliable patterns in the history of incumbents, and it deserves a name, because it is about to repeat at every company whose product is an interface.

And it isn't only his word that this landed. Within a day the post had been seen well over two million times and picked over on Hacker News,[7] and another well-known developer — one who maintains a rival Workspace tool of his own — summed it up for a few hundred thousand people: Google fired the guy that made the Google Workspace CLI, because he made the Google Workspace CLI.[8] The deeper claim isn't his invention either. Months earlier, a founder who'd spent a decade building interface-heavy software wrote the same thing from the other side of the table: the interface wasn't a feature; it was the product — and once an AI agent becomes the interface, the years of muscle memory that once locked customers in simply dissolve.[9] The firing is one man's account. The pattern is not.

Organizations do not miss disruption. That is the comfortable story — that Kodak just didn't see digital coming, that Blockbuster wasn't paying attention. They saw. The harder truth is what an organization does in the moment it sees: it attacks the evidence of the disruption, because the evidence is the only part it can reach.

Strip the drama and look at the artifact. A command-line tool is not exciting. What made this one climb Hacker News was the thing it quietly demonstrated: that Workspace — the menus, the buttons, the decades of accreted interface — is, from an agent's point of view, overhead. An agent doesn't want a place to click. It wants an instruction it can issue and a permission to issue it. Collapse the product to that, and most of what you can see of Workspace turns out to be packaging.

That is not a feature critique. It is an existential one, and everyone in the building can feel the difference. A feature critique says your product could be better. This said your product could be a thin shell around data an agent can reach on its own — a wrapper, in the jargon — and a motivated engineer can prove it in a weekend. The tool worked. That was the problem. You can survive a competitor building the cheap version of your product. It is harder to survive one of your own people building it, publishing it, and watching the internet agree.

Confronted with that, an organization has a targeting problem. It cannot fire the trend. It cannot escort “agents are coming for the interface” out of the building; there is no one to walk to the door, no commit to revert. But the demo is concrete. The person is concrete. And a threatened system reaches for what it can actually grip.

This is Clayton Christensen's innovator's dilemma — the finding that great companies fail not despite doing everything right but because they do, listening to their best customers and defending their best margins until the cheap, worse-looking thing grows up and eats them.[5] We usually meet the dilemma as a chart: market share, an incumbent's line crossed by an upstart's. But from the inside, in real time, it doesn't look like a chart. It looks like a personnel decision. Disruption arrives first as an embarrassment — a demo that makes the flagship look slow — and the cheapest way to make an embarrassment go away is to remove the person holding it.

Xerox learned this the expensive way. Its Palo Alto Research Center invented the graphical interface, the mouse, the windowed screen — the entire vocabulary of modern computing — and Xerox, a company that sold copiers, let it walk out the door to Apple, because the people who ran Xerox could see that the future PARC was building did not look like the business that was paying them.[6] Kodak built a working digital camera in 1975 and shelved it, because the prototype was an argument against the film that was the entire company. In each case the institution did not fail to see. It saw with perfect clarity, and chose the thing that was paying it now.

I'd call this the Antibody Response. The metaphor is the immune system, which exists to do one job: detect something that doesn't belong and destroy it before it spreads. That machinery is what keeps you alive. It is also, in an autoimmune disorder, exactly what turns on the body it was built to protect — precision aimed at healthy tissue. An organization's immune system works the same way, and fails the same way. It runs in three beats.

1. 01.

   The Trigger

   An insider makes the disruption concrete and public. Not a strategy memo, not a slide that says “agents are a risk” — a working thing people can use and love. Abstraction is survivable; a demo is not. The trigger isn't the threat itself. It's the threat made undeniable, on a Tuesday, with a star count climbing in public.

2. 02.

   The Misfire

   Unable to attack the trend, the immune system attacks the carrier. It mistakes the proof for the pathogen. The reasoning is never spoken as “destroy the evidence” — it arrives as something procedural and plausible: a branding violation, a process not followed, a reorg, a role no longer needed. The antibody doesn't feel like fear from the inside. It feels like policy.

3. 03.

   The Cost

   It does not slow the disruption — Google announced its own CLI regardless; the agent future arrives on schedule. What the response actually destroys is the organization's ability to learn from its own early signals, and the lesson every other employee quietly absorbs: don't build the dangerous demo, don't be the one who proves the uncomfortable thing. The antibody response doesn't cure the disease. It blinds the host to the next one.

Which leaves two people with something to do on Monday, depending on which one you are.

If you are the builder, understand the position you occupy the moment your side project starts working. Inside an incumbent, being right too early and too loudly does not read as initiative. It reads as an antigen. That is not a reason to stop building — it is a reason to be clear-eyed about where the proof is worth the most. The demo that gets you removed from one company is often the demo that starts the next one. Xerox's interface became Apple. The proof travels better than the person who carries it — but it travels best when the person carries it out the door.

If you are the leader, the internal demo that embarrasses your own product is the single most valuable artifact your organization will produce this year. It is free competitive intelligence, delivered by someone who still works for you and still wants to. The urge to make it quiet — to find the branding violation, to question the process — is the disease presenting as good judgment. When you feel the impulse to fire the messenger, you have just located your own blind spot with perfect accuracy. The correct move is the hard one: fund the thing that scares you, and put the person who proved your moat was thin in charge of what comes after it.

• Find your most uncomfortable internal demo. The side project that makes a core product look slow is not a morale problem — it's a forecast. Read it as one.
• Separate the threat from the carrier. Before anyone is reorganized or declared “not a fit,” ask out loud: are we removing a problem, or removing the evidence of one?
• Reward the proof, not the comfort. The person who shows you the cheap version of your own product did you a favor at their own expense. Pay it back — or everyone else learns not to.
• Assume the trend survives the firing. Whatever you make quiet internally is still true externally. The only thing the antibody changes is whether you hear it first from your own people or from a competitor.

Disruption is rarely a failure of vision. The incumbents almost always see it coming. It's a failure of immune regulation — an organization, working exactly as designed, attacking the cells that were trying to save it. The agent layer is going to dissolve a lot of interface moats. Watch who gets fired. They're the early-warning system, and the firing is the alarm.