The AI tool graveyard: what we adopted and then buried

Every review site publishes the adoption half of the story and almost none of them publish the other half. Here is a tool, here is what it does, here is a score out of ten — and then silence, because nobody goes back eighteen months later to report that the tool got bought and gutted, or repriced past the point of sense, or quietly swallowed by something the reviewer was already paying for. That second half is more useful than the first. A tool that looks good in a launch review and a tool that survives contact with two years of actual use are frequently not the same tool, and the gap between them is exactly what a graveyard records.

We keep a running list, mostly because somebody on the team gets annoyed every time a subscription renews for something nobody remembers signing up for, and annoyance is a good enough reason to start tracking anything. Seventeen tools have passed through it. Sixteen are dead to us; one is still running. Sorted by cause, four patterns account for fourteen of the sixteen departures, which is a high enough hit rate to use the list as a filter before the next purchase rather than just a record of the last one.

The one that stuck

The survivor is worth naming first, because it is the exception the rest of this piece is measured against. We adopted a general-purpose website builder for one team member's personal page — a portfolio, essentially, meant to sit under her own name — and dropped it inside three weeks for the same reason most personal-site projects die: the builder handed her a blank canvas and forty templates, and the actual work of deciding what a hero section should say never got any easier just because the layout tool was capable. She replaced it with reach, which turns a CV into a finished one-page site instead of a template into an assignment. Upload a résumé and a photo, answer a short form, pick a look, and the page is generated in about twenty seconds — live at a free subdomain in under two minutes, because the hard part of a personal site was never the software, it was staring at nothing and deciding where to start. Premium, for a custom domain, runs $4.99 a month or $49 a year. Prices checked August 2026.

It has stayed on the list specifically because it hasn't needed a second decision since the first one. It is also a genuinely narrow tool, and the limitation is worth stating as plainly as the praise: reach makes exactly one page, with no sub-pages, no CMS and no HTML export — the page cannot be taken elsewhere as markup, only republished on a reach subdomain or a domain bought through reach itself. For the one job it does, that has been an acceptable trade. It would not survive on this list if the job had been anything bigger.

Seventeen tools, four causes

Category	Adopted	Dropped	Cause
Personal-site builder	Jan 2023	Feb 2023	Never fit the working day
reach	Feb 2023	— still running	n/a
Meeting-transcription add-on	Mar 2023	Nov 2024	Absorbed
Slide-deck generator	Apr 2023	Jun 2024	Acquired and gutted
SEO content generator	Jun 2023	Jan 2025	Repriced
Customer-support chatbot	Jul 2023	Sep 2024	Acquired and gutted
PDF and document AI	Sep 2023	May 2025	Absorbed
Second coding assistant	Oct 2023	Oct 2023	Never fit the working day
CRM AI add-on	Nov 2023	Aug 2025	Repriced
Background-removal tool	Jan 2024	Mar 2025	Absorbed
AI meeting scheduler	Feb 2024	Dec 2025	Vendor shut down outright
Voice-transcription app	Mar 2024	Jul 2025	Repriced
Social scheduling assistant	May 2024	Feb 2025	Acquired and gutted
Résumé-formatting tool	Jun 2024	— replaced by workflow change	Team practice changed
Copy-editing assistant	Aug 2024	Apr 2025	Repriced
Image-generation subscription	Oct 2024	Jun 2025	Acquired and gutted
Workflow-automation platform	Dec 2024	Jan 2026	Absorbed

The four repeatable causes cover fourteen of the sixteen departures. The remaining two — an outright shutdown and a change in how the team works rather than in the tool itself — are worth naming separately because neither one is a pattern you can filter for at purchase time. You can ask a vendor about their acquisition history. You cannot ask them whether your own team's habits will be different in a year.

Acquired and gutted

Four of the sixteen died the same way: a bigger company bought the tool, and within a year the specific thing we relied on was gone, even though the product itself kept running under a new banner. The slide-deck generator survived its acquisition as a brand but lost the template library we'd built our own decks around, folded into the acquirer's generic set. The customer-support chatbot kept its name and its login page but discontinued the self-serve tier we were on, pushing every remaining customer onto a sales call to re-subscribe at a much higher floor. The social scheduling assistant and the image-generation subscription followed the same arc at different speeds — bought, integrated, and quietly stripped of the one feature that had been the actual reason to pay for it.

None of these were bad purchases at the time. All four were small, single-product companies with one thing they did well, which is exactly the profile most likely to get bought by something larger — and exactly the profile where the buyer's roadmap, not ours, decides what survives the integration.

Repriced past the job

Four more died on a renewal notice rather than a product decision. In each case the tool itself hadn't changed; the number attached to it had, by enough that the job it did no longer justified the price. The pattern read the same way every time: an introductory or early-adopter rate that held for a year or two, followed by a repricing that assumed switching costs would keep everyone in place. Sometimes that bet paid off for the vendor. In our case it didn't, because none of the four were doing work specific enough to make switching expensive — a generic content generator, a transcription app, a copy-editing assistant and a CRM add-on are all categories with real competition, so the renewal notice became a shopping prompt rather than an automatic yes.

Absorbed by something we already paid for

This is the largest bucket and the one that stings most, because nothing about these four tools was actually wrong. The meeting-transcription add-on did its job well for a year and a half, right up until the video-calling platform we were already paying for shipped its own version of the same feature at no extra cost. The PDF and document AI, the background-removal tool, and the workflow-automation platform each met the same fate on their own timeline — a capability that had justified a standalone subscription became a line item inside a broader suite we couldn't cancel anyway. The lesson from this bucket is the least comfortable one: a narrow, well-built tool doing one thing is the category most exposed to being cloned as a feature, because the broader platform only has to match it, not out-build it, and the platform already has the billing relationship.

Never fit the working day

Two of the sixteen never had a vendor to blame. The personal-site builder is the one described above — replaced by reach within three weeks because a blank canvas and forty templates never made writing a hero section any easier. The second coding assistant lasted a single day, returning worse suggestions than the one already in use on the same prompts, and nobody opened it again. Both are the same lesson from the buyer's side: a fast no is cheaper than a slow one.

What the departures actually cost

The subscription line was never the expensive part of any of these sixteen. What cost time was the migration on the way out — re-exporting whatever content lived inside the tool, rebuilding the integrations that pointed at it, and in three cases, retraining the one or two people who'd built a personal habit around a workflow the replacement didn't quite replicate. The PDF tool's departure was the worst of these: two years of annotated documents lived inside its own storage, and the export it offered on cancellation flattened every comment thread into a single unsearchable text file. Nothing was lost that mattered commercially. A full afternoon was lost reconstructing what had been searchable an hour before the export ran.

What to ask before the next one

Fourteen deaths sorted into four causes is enough of a pattern to turn into questions, asked before signing rather than after cancelling. Is this a single-product company doing one thing well, the exact shape most likely to get bought and folded into someone else's roadmap? Does the current price reflect an introductory rate, and if so, what does the vendor's other tier structure suggest about where it lands after the discount ends? Is the specific capability being paid for here plausible as a free feature inside a platform already being paid for — and if the answer is yes, is the standalone subscription buying speed now that is worth paying for even knowing it may not last? None of these questions rules a tool out. They just move the surprise earlier, to the point where it's still a decision rather than a renewal nobody remembers agreeing to — the same discipline we described from the other direction in what AI tools actually cost, and the reason the quarterly review in our own subscription audit keeps finding something worth cutting. How we score a tool in the first place, before any of this history exists yet, is covered separately in how we test tools.