Nest bricked "lifetime" smart home hubs and paid refunds. Here is how to kill a feature without that bill

Compute budgets are tightening. AI tokens cost real money now. And that old feature nobody wants to touch is still quietly billing your team for engineering hours, support tickets, and compliance work. You feel it in every roadmap review when there is no room for the new bet. Let me catch you up on how to cut the dead weight without burning the customers who still depend on it.

The deep cut

• A feature you would not fund today is already dead. Dean at Citrix kept bundling GoToMyPC years after end of sale because no one made the call.
• Talk to legal and finance first, engineering last. Nest bricked Revolv's "lifetime" hubs and drew an FTC refund demand.
• A slow deploy path is a quality filter, not a tax. Gothelf warns that new builders will route around gates and let customers do your QA.

The feature nobody owns

A zombie product has stopped creating value but keeps eating resources, because nobody will say the word. Dean at Productside got the tap on the shoulder: "Dean, please bring GoToMyPC to end of life." The rest of the GoTo line was sold off. Sales kept bundling GoToMyPC into deals anyway. It shuffled along for years.

Run a simple test on each feature. Does it still eat capacity? Would you fund it today if it showed up as a fresh pitch? Does anyone own the decision? If the answers are yes, no, and no, you have a zombie. Decline is a normal life stage. A zombie is a failure to decide.

The cost you miss is the one nobody writes down. Kodak invented the digital camera in 1975, then let film revenue hold it back for years. An agricultural drone team Dean worked with could not ship better, cheaper units while the factory kept building the old ones. The bill is the next bet you cannot place.

Why AI makes the pruning urgent

The old forces that create zombies still apply: inertia, sunk cost, politics, legacy contracts. AI adds new ones. Dean wonders how many "feral agents" companies have running around since last year, burning tokens with nobody watching the meter. The seagull manager who vibe-codes an agent over a weekend and leaves the team to keep it alive is now a real line item.

This is the same chaos Jeff Gothelf flags when everyone becomes a builder. More things get shipped, fewer get owned. Sonia Vora, a former Apple PM, pushes her PMs and designers to push their own PRs as long as a strong code review process catches the mess. Speed up the front of the pipe and your zombie count climbs at the back.

Instrument down to the feature level. The data tells you which features are not earning their keep, so you can cut a bad flow without killing the whole product. Run the check quarterly, monthly for anything fast-moving.

Killing it without stranding the customer

The one rule: lose the legacy product without losing the legacy customer. When Nest shut down the Revolv smart home hub, it turned devices sold with a "lifetime" subscription into paperweights. The FTC looked into it and Nest ended up offering refunds. A proper sunset would have cost far less goodwill. Dean expects more of this as cash-burning AI companies get acquired or shut down and leave customers with little warning.

Talk to legal and finance first, engineering last. Contracts, bundles, and projections all change when a product retires, and you need to know what you promised before you touch anything. Then customer success, then sales and partners, then a few key accounts, then engineering. When Dean brought GoToMyPC to end of life, the Citrix lawyers already knew him by name, because the first time they met was not a favor.

Give customers a path, not a goodbye email. For the GoToMyPC accounts, Citrix moved them to the buyer running the same codebase, with a discount and no data migration. Kenny kept a large financial services client by giving them priority access to the replacement and building their must-have features into the first versions.

Make the gate fast enough to use

If you let designers and PMs ship, redesign the path they ship through. Gothelf's framing is clean: the builder is the user, the deployment process is the product, and how well people pass through it is your outcome. The gates exist to catch weak ideas before customers do. You do not want fewer gates. You want people to move through them faster.

Set two time targets. Compliant code clears all reviews in hours. Non-compliant changes get actionable feedback in minutes, not silence or "I don't like it." Automate the deterministic checks: tests pass, security scans clean, no new unowned dependencies. Reserve human review for the hypothesis and the kill-or-keep decision.

Every shipped feature should carry a named owner and a review date for a kill, pivot, or persevere call. That is your anti-zombie insurance. Build the off-ramp at launch and you stop breeding the next feature nobody wants to touch.

Three questions for your team

• Which feature in our portfolio would get zero funding if it were pitched fresh today, and who owns the call to retire it?
• Before we sunset anything, have we mapped every contract and bundle commitment with legal and finance, or are we about to strand a customer like Nest did?
• Does every feature we ship this quarter have a named owner and a review date, or are we minting zombies faster than we kill them?