Ship the backlog
without hiring another engineer.

Both, in that order. Agents write the first version: spec, failing test, implementation, screenshots, PR notes. I read the diff, the test, the screenshots, and the PR description before any of it reaches your queue. The model gets faster every quarter; my judgment is what stays the same.

You do not merge it. Comment what is wrong, the pipeline reworks it (or I do, depending on the issue). Same as any other code review. PRs only land on your main branch when you click the merge button.

Three things: (1) I do the planning step that turns a one-line GitHub Issue into a real spec the agents can run against. (2) I sustain it 24/7, not just when you have time to babysit a chat window. (3) Every PR gets a senior review before it touches your branch. You can absolutely build this loop yourself. I am selling the time you would spend doing it.

Yes, with caveats. The pipeline does best in repos where conventions are at least partially documented (a CLAUDE.md, a README, or a few good test files signal what good looks like). The first kickoff call covers what is messy and what to ignore. If your repo is so legacy that there are no tests at all, the first few items will run slower while we lay groundwork.

The continuum-backlog bot account needs collaborator access with: read repository, push to feature branches, open pull requests, comment on issues. It does NOT need (and will not be granted): admin, settings, secrets, write access to main. Your branch protection rules stay intact.

Yes, on the Plus and Pro tiers. The architect agent posts the implementation plan as an issue comment before work begins. You can approve, request changes, or reject. On Starter the bot proceeds after estimation unless the plan exceeds your tier cap, in which case it stops and asks. If you want plan-approval on Starter, it slows turnaround but is fine.

You own the code. The pipeline pushes commits to your repo under the bot account; on merge they are part of your codebase, same as any other contributor. We do not retain copies and we do not claim rights. AI-generated code is treated as derivative of the agent's training, which is consistent with how courts have treated this so far. If you have a specific license or attribution concern, surface it on the kickoff call and we can decide together what fits.

You catch it at one of three places. First, the Playwright test will fail before the PR opens. Second, my review catches what the test missed. Third, the screenshot diff shows visual regressions you would have missed in code review. If something still gets through, you do not merge it. Worst case, you merge a bad change and revert it. We are not bypassing your normal merge process; we are feeding it.

Yes. Self-serve from your Stripe billing portal (the link is in your welcome email). Cancellation kicks in at the end of the current billing period. 7-day money-back guarantee: cancel within 7 days of your first charge or any subsequent monthly renewal and you get a full refund — no email back-and-forth required.

Your code is sent to model APIs for inference (this is how the agents read context and generate diffs). It is not retained or used for model training under our agreements with Anthropic. Code never leaves your repo permanently. The pipeline pulls a sandboxed worktree on my server, runs the work, pushes the PR, then cleans up. NDA available on request, no extra charge. If you need fully on-prem (no inference calls leave your network), that is not something this product supports today.

First PR Sprint is 48-72 hours. Starter has no fixed SLA (work picks up when the queue clears). Plus is 5-day pickup. Pro is 48-hour pickup. Most small fixes open as PRs within one business day after pickup. Mid-sized features take 2-4 days. Multi-PR Pro features are scoped per feature. The architect agent posts an ETA in the issue comments before work begins so you are never guessing.

The architect agent estimates scope before work begins. If an issue exceeds your tier's scope, it comments back with three options: split into smaller issues, upgrade to the next tier (prorated), or pay a one-off Sprint. No silent overruns and no work happens without you knowing.

Realistic concern. I run this solo and that is the trade-off. If I am away (illness, holidays, travel), I tell you upfront and credit the days back on your next renewal so you do not pay for days when no work is happening. The pipeline can keep specing and drafting PRs while I am away, but they queue waiting for review and do not reach your branch until I am back. If unavailability becomes a recurring concern, that is the signal to move to a partnership-shaped offer instead, and we restructure.

Yes, that is the recommended path. Book a free 30-min backlog review. We walk through your repo, your backlog, and figure out which tier fits. If a tier fits, you can subscribe on the call. If it does not, I tell you and we figure out the next step.