Sometimes, yes, you should build it yourself. A coding agent genuinely compresses the scaffolding, the boilerplate and the first working version of a feedback portal. It does not compress deciding what the thing should do, reviewing what got written, security, or being the person paged when it breaks.
Canny.io published its own costing on 19 August 2026. The figures below are theirs, quoted and attributed, not ours.
What did Canny.io actually price out?
On 19 August 2026, Canny.io published a piece costing a do-it-yourself feedback portal, feature by feature. Twenty-five features, priced line by line, landing on roughly 420 engineering hours and about $31,500, plus running costs rising from $45 a month to $550 a month at the compliance end. It names Claude Code directly, which is a specific, checkable claim rather than a vague swipe at "AI coding tools" in general.
Their argument, in their own words: "the code was never the hard part." The hours go to review, testing, permissions and the syncs between systems, not to typing out the screens.
Is that number actually fair?
Largely, yes, and it is worth saying plainly: this is the most honest build-versus-buy costing anyone in this category has published. It is also, conspicuously, the only cost comparison Canny.io runs, and it runs in the one direction that flatters a subscription. Both things can be true at once.
What does a coding agent actually compress?
This is the part worth separating out, because "a coding agent changes everything" and "a coding agent changes nothing" are both wrong in ways that matter to your decision.
| What a coding agent compresses | What it does not |
|---|---|
| Scaffolding and the first working version | Deciding what the thing should actually do |
| Boilerplate: forms, routes, database schema | Reviewing what the agent actually wrote |
| The long tail of small features nobody gets around to | Security: authentication, permissions, injection, rate limits |
| Wiring third-party services together | The ongoing maintenance, month after month |
| Redoing its own work once you tell it what is wrong | Being the person paged when it breaks |
So has Canny.io's estimate gone stale?
Not in the way that would change its conclusion. The hours Canny.io counted were never mostly typing; by their own account they went to review, testing, permissions and the syncs. Those are exactly the hours a coding agent does not remove, because they are judgment calls about somebody else's data, not code that needs writing faster. What shrinks is the hours before that: getting from nothing to a working first version, and the pile of small, unglamorous features that used to be the reason a self-build dragged on for months. That is real, and it is not nothing. It is also not the 420 hours.
What does it cost to keep the thing running, not just build it?
Building it is only half the bill. UserJot, arguing against self-hosting, publishes the more honest running cost: a self-hosted Fider deployment costs $70 to $125 a month in infrastructure before any engineering time, split across Postgres, hosting, transactional email, and monitoring and TLS, plus five to ten hours a month once traffic is real. A coding agent does not touch that figure at all. It has no opinion on your Postgres bill, and it is not the one who gets paged when the email provider changes its API.
When should you actually build it yourself?
- You enjoy this kind of project, and the time is going somewhere you would have spent it anyway.
- Your requirements are genuinely yours alone, not a shape any packaged tool fits.
- You were going to run infrastructure regardless, so the marginal maintenance is close to nothing.
- The tool is small enough that "ongoing maintenance" is realistically an afternoon a year, not a standing job.
When should you not?
- If what you are actually avoiding is a subscription rather than a genuine mismatch with what is on offer. A subscription is a known monthly number. A self-built tool trades that for an unknown one, paid partly in money and partly in your own time.
- Your own time is the least visible line in any budget. It does not arrive as an invoice, so it is easy to spend without noticing, and easy to underprice when you are deciding afterwards whether building was worth it.
One payment, no subscription, unlimited products.
Where does buying it once fit in?
There is a third option: pay once for something already built, maintained by someone else, and move on. Docket is that option here, one payment, no subscription. It is not automatically the right answer. If you would enjoy building the thing, or your requirements genuinely do not fit anything packaged, building it is a perfectly good choice, and a coding agent makes that choice faster than it used to be. The honest reader of Canny.io's own costing, and of this page, is someone who can make that call with clear eyes rather than being talked into either side of it.
For the full three-year comparison across every tool in this category, see what feedback tools actually cost. If you are weighing self-hosting more broadly, self-hosting a feature request board covers the wider case, and static site or server covers the architecture question underneath it.
Frequently asked questions
How much did Canny.io say it costs to build a feedback portal yourself?
Roughly 420 engineering hours and about $31,500, plus running costs rising from $45 to $550 a month at the compliance end, across twenty-five features priced line by line. Canny.io published the costing on 19 August 2026 and names Claude Code directly.
Does a coding agent make Canny.io's estimate too high?
Not by as much as it might look. Canny.io's own account puts most of the hours into review, testing, permissions and the syncs between systems, which a coding agent does not remove. It does compress the scaffolding, the boilerplate and the first working version, which is real time saved, just not most of the 420 hours.
What does it cost to run a self-hosted request board, separate from building it?
UserJot, publishing figures to argue against self-hosting, puts a self-hosted Fider deployment at $70 to $125 a month in infrastructure before any engineering time, plus five to ten hours a month once traffic is real. That cost exists whether or not a coding agent wrote the code.
Should I build my own feedback tool or buy one?
It depends on what you are weighing. Build it if you would enjoy the project, your requirements are genuinely unusual, or you were already going to run infrastructure. Do not build it if the thing you are actually avoiding is a subscription rather than a real mismatch, since your own time is the least visible line in any budget.
Is Docket cheaper than building it yourself?
On the numbers Canny.io published, yes: $99 once against roughly $31,500 in engineering time plus $45 to $550 a month to run. But that comparison only matters if a packaged tool actually fits what you need. If it does not, the cheaper option on paper is not the right one.