Should I vibe code
Project workspace with tasks, documents, automation, and AI coordination
Height shut down in September 2025 and deleted the data 30 days later. So much for just buying the SaaS.
?
Their verdict, the Core (final list price) price and the build-time estimate come from their entry, MIT-licensed. Checked 2026-08-04.
?
Our verdict, the regret score and everything below it. Editorial and unsponsored — nobody can pay to be moved.
The honest answer
why the verdict is what it is
Height is the entry where the argument makes itself: the product shut down on 24 September 2025, six months after the announcement, with customer data deleted thirty days later. 'Just buy it' is not on the table, so the only question left is whether to build. A task list is honestly an evening. A multiplayer workspace — realtime sync, offline edits, notifications, guests, search across a year of tasks — is not, and the autonomous half Height was betting the company on is the piece to be most careful with: an agent that de-duplicates, renames and closes tasks is a destructive write path over the only record of what your team agreed to, running while nobody is watching.
What actually breaks
not "if". the specific failures.
- Realtime sync, the first time two people edit one task offline and your merge rule turns out to be last-write-wins
- The autonomous half: auto-triage closing a stale task somebody was waiting on, with no audit row naming what did it
- Deduplication, which is a destructive operation wearing tidying's clothes
- Notifications, the feature that makes a tracker useful and quietly turns your domain into a bulk sender
- Search, which is fine at two hundred tasks and useless at twenty thousand
- Guest permissions, the day a client gets an account and can see the list next to their own
- The export, which nobody writes until the week they need it — ask anyone who was using Height in March 2025
Is that you?
the verdict is a default, not a law
- It is your team, your data, and there is a scheduled export you have restored from at least once
- The AI suggests and a human applies — no unattended writes to shared state
- Task history is append-only, so 'who changed this, and why' has an answer
- You are replacing a spreadsheet rather than replacing Linear
- Clients or contractors get accounts and you are writing the permission model yourself
- An agent may close, merge or rename tasks unattended
- The only record of what the team agreed lives in one database with no export
- You need offline editing, because CRDT-shaped problems are not weekend-shaped problems
If you build it anyway
the checklist, then the prompt that enforces it
- Write the export first: newline-delimited JSON of every task, comment, attachment and audit row, on a schedule, into storage you control. Height's customers got six months of notice — your side project will give you none.
- Make the task log append-only. Every mutation carries an actor, a timestamp and a reason, so an AI-driven change is distinguishable from a human one and both are reversible.
- Do not let a model write to shared state unattended. Suggestions go into a review queue and a human applies them. 'Autonomous' is a marketing word for unattended destructive writes.
- Soft-delete everything with a restore path and a tombstone. In a tracker, deletion is almost always somebody else's mistake.
- Decide the conflict rule before building any offline support and write it down — field-level merge, last-write-wins with a visible warning, or no offline editing at all. Silent loss is worse than a refused edit.
- Send notifications through a transactional provider, batch them, and give every user an off switch, or your tracker becomes the reason your domain gets throttled.
- Index for search from day one and test it against a synthetic year of tasks. Search is the feature that decides whether the tool survives its second quarter.
I am building a team project tracker: tasks, lists, comments, and some AI
assistance. Apply these constraints and push back if I ask you to break them.
1. Build the export before the board view. A scheduled dump of every task,
comment, attachment and audit row as newline-delimited JSON plus files, into
storage I control, with a documented restore. Tell me to run the restore once
before the team starts using this.
2. Every mutation is an append-only audit event with actor, timestamp, before and
after. No in-place updates without a corresponding event.
3. Anything the AI proposes goes into a review queue for a human to apply. Do not
build unattended agents that close, merge, rename or reassign tasks. If I ask
for autonomous behaviour, implement it as a dry run that writes a report first,
and make me read one week of reports before it can act.
4. Deletion is soft, with a tombstone and a restore path, including bulk actions.
5. Decide and document the conflict rule before any realtime or offline work. If
I ask for offline editing, tell me plainly what CRDTs cost to get right and
offer the simpler option of refusing stale writes.
6. Notifications go through a transactional provider, are batched per user, and
have a working unsubscribe. Never send from an application SMTP loop.
7. Permissions are checked per record on every read and write, not per page, and
guests are a separate role with an explicit allowlist of what they can see.
8. Build search over task text and comments early, and test it against a
generated year of data rather than a demo project.
9. Keep integrations one-way in v1 — read from GitHub or chat, do not write back.
Two-way sync doubles the surface and quadruples the conflicts.
10. Out of scope: documents, chat, dashboards and roadmaps. Each is a separate
product, and the tracker will not be finished before you get there.
That one keeps you out of trouble. For the prompt that actually builds it, canivibecodeit.com has one.
their build prompt ↗Or don’t build it
the boring option, and the way back out
You cannot — Height is gone. If Height is what you wanted, the closest live equivalents are Linear, which we rate DEMO ONLY at around $10 a seat, or self-hosted Plane if you would rather own the box. The useful lesson is the one the shutdown handed everybody for free: choose the tool whose export you could actually use, and try it before the email arrives rather than after.
One line-delimited JSON file per table, attachments as files with a manifest, dumped nightly somewhere that is not the app's own disk. It is worth being blunt about the standard here: the vendor gave six months' notice and then deleted everything, and that is the good version. Your own project ends the weekend you lose interest, and the export you never wrote is the difference between moving to Linear in an afternoon and retyping a year of decisions.
Open-source task manager with lists, boards, teams and a documented API.
Open-source issue and project tracker, self-hostable and the closest thing to a drop-in replacement.
Questions
Height really shut down?
Yes. The founder announced it on 20 March 2025 and the final day of service was 24 September 2025, with a commitment to delete customer data within thirty days of that date. The last archived version of height.app is the farewell letter rather than a product page, which is why this entry has no current price and why the 'just pay for it' argument is unavailable.
Does a dead vendor make building your own more attractive?
A little, and it is worth being honest about how much. It removes the assumption that the commercial option is the durable one — six months of notice is generous by the standards of venture-funded tooling. It does not make realtime sync, permissions and search any easier. The right conclusion is not 'build your own tracker'; it is 'weight the export, not the feature list, when you choose'.
What about the autonomous AI features specifically?
That is where the risk sits in a build like this. Suggestions in a review queue capture most of the value: a proposed duplicate merge, a proposed triage, a proposed close. Unattended writes capture the rest and cost you the ability to answer 'why did this task close'. If you build one thing from this page, build the audit log before you build the agent.
Every week, someone ships something they shouldn’t have.
New verdicts, the worst thing that landed in the trap, and the occasional incident report. No other email, ever.
A no-code platform is not an app, it is a type system with a formula engine bolted on. You are building a compiler.
A search box says "here are three pages". An answer box says "this is true". Only one of those can be wrong.
The board is a weekend. Being the reason nobody can file a bug on Monday is the rest of your year.
last reviewed 2026-08-05 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice