# You have the idea. You don't have an engineer.

File a task in plain language. CodeHerder's agents plan, code, review, and merge it for you.

Source: https://codeherder.com/ship-without-an-engineer/

For product people & non-technical builders

Write the task the way you'd write it to a teammate: what's wrong, or what you want. An agent plans it, builds it, then opens a pull request while you wait.

[Request access →](https://codeherder.com/get-started/#request-access) [See what people actually write →](https://codeherder.com/ship-without-an-engineer/#what-you-write)

Whole-lifetime cost of the five plain-English tasks in /examples#nocode.

What you actually write

## Three real tasks, filed exactly like this

No ticket template, no required fields. These are the actual sentences three people typed into the web app, and what came back.

- "Missing skills link in Set Up section on Group"Added the missing nav link, so a Skills workspace a group inherits is reachable from that group's own Set up section. $3.05 · 20m [Read the full story →](https://codeherder.com/examples/skills-link-in-groupnav/)
- "Drop down arrow cut off on mobile. There is now also no way to get to your profile on mobile."Fixed the mobile topbar overflow, which also restored the only path to Profile, Theme, and Sign out on a phone. $20.87 · 4h 53m [Read the full story →](https://codeherder.com/examples/mobile-usermenu-cutoff/)
- "We recently changed it so that devices didn't keep the names of the ec2 servers they were created from - probably an ok change... but what it has been replaced with is useless... some hash that means nothing..."Added an operator-set device name that now wins everywhere a device is shown, with the old hostname kept as a fallback. $29.54 · 3h 9m [Read the full story →](https://codeherder.com/examples/device-display-name/)

Two more like these, all filed by people with no CLI and no git, at [/examples#nocode](https://codeherder.com/examples#nocode). If a word like [task](https://codeherder.com/glossary/task) or [orchestration](https://codeherder.com/glossary/agent-orchestration) is new to you, the [glossary](https://codeherder.com/glossary) defines it in plain English. Looking for a specific job instead? See [ship a small feature](https://codeherder.com/use-cases/ship-a-small-feature) and [clear the bug backlog](https://codeherder.com/use-cases/clear-the-bug-backlog). Want a ready-made starting point instead of a blank page? Copy one from [starter briefs](https://codeherder.com/starter-briefs).

What happens after you hit save

## The pipeline runs without you. You show up twice.

Once a task leaves your inbox, it works through six stages: plan, code, review, merge, verify, done. Agents do the work in five of them. You're asked for something at two moments along the way.

### You approve the plan

Before any code exists, an agent writes out what it's going to do and why. Read it, and either let it continue or send it back with a note.

### Everything in between runs on its own

Code → review → merge happen without you watching. A separate review stage checks the work before it's allowed to merge, so one agent doesn't grade its own homework.

### You accept the result

Verify confirms the merged change actually does what the task asked for, then hands you a plain-language summary. Accept it, or say what's still wrong.

What you need before day one

## Three things, and only one of them is technical

- A repository Your code, wherever it already lives. CodeHerder never becomes a host for it.
- One connected machine A laptop or a small server works. Agents run there, in their own isolated copy of your repo.
- About an hour of a technical person's time, once Connecting that first machine is the one step that needs a developer or your IT team. Borrow one for an hour, or see [where agents run](https://codeherder.com/where-agents-run) for a one-click option that needs no CLI at all. After that, it's a browser job.

[See the exact setup steps →](https://codeherder.com/get-started/#path-a)

How the work gets checked

## Four gates, and none of them is optional

- Plan You read what the agent intends to do before it writes a line of code.
- Review A second, separate stage checks the code itself. The agent that wrote it never grades its own work.
- Merge Blocked until review passes. Nothing reaches your repository's default branch on its own say-so.
- Verify Confirms the merged change does what the task asked for, and reports back to you.

More on how each gate is enforced at [/verification](https://codeherder.com/verification), and on where your code and keys stay at [/security](https://codeherder.com/security).

What it costs

## What five real tasks cost, start to finish

Every figure here is the whole-lifetime cost of the five plain-English tasks in [/examples#nocode](https://codeherder.com/examples#nocode), not an estimate.

$3.05 Cheapest $21.75 Median $29.54 Most expensive

Estimate your own workload at [/calculator](https://codeherder.com/calculator), or see the plans at [/pricing](https://codeherder.com/pricing).

FAQ

## Questions people ask before filing their first task

Do I need to know how to code?

No. You write the task the way you'd write a message to a teammate: what's wrong, or what you want, in plain language. There's no CLI and no git anywhere in the web app.

Who fixes it if the agent gets something wrong?

You see the plan before any code is written and the result before it merges, so you can send either one back with a note. A separate review stage checks the code too, and a merge gate blocks anything that fails it.

Do I need my own repository?

Yes: one repo and one connected machine, set up once by a developer or your IT team. See /where-agents-run for the setup options, including one with no CLI install. After that, every task you file runs against your own code, in its own isolated copy.

Can I try one small thing first?

Yes. File a task like a broken layout or a missing link and watch it go from a sentence to a merged fix. /examples has real ones, filed by people with no CLI, cost included.

What happens to my source code?

It stays in your own repository the whole time. An agent works on a copy on your connected machine and opens a pull request back to that repository when it's done; your code never passes through CodeHerder itself.

What if I disagree with the plan?

Say so before it turns into code. The plan stage is where you push back: leave a comment or send it back for a rewrite, and the agent revises before a single line changes.
