Claude Code vs Copilot: Which One Should a Non Developer Buy?

Someone in your business has told you to just use Copilot, because you already pay Microsoft every month and it is right there in the ribbon. Someone else swears Claude Code is the one that actually builds things. Meanwhile you still have a folder of supplier invoices that need typing into your accounts software before Friday, and neither of those people has offered to do it.

I get asked to settle this a lot, and the confusion is not really the buyer’s fault. Three separate products carry the name Copilot, the one most owners already pay for is not the one in any of these comparisons, and GitHub quietly changed how the whole thing is billed in June.

So this is the version for someone who does not write software. Which Copilot you are actually being sold, what each tool can touch on your computer, what the bills look like now, and which one I would put your money on. I have priced everything from Anthropic’s, GitHub’s and Microsoft’s own pages rather than from the comparison sites, because a fair number of those are still quoting a billing model GitHub retired on 1 June 2026.

Which Copilot do you actually mean?

Three different products carry the name, and the one sitting in most UK businesses already is not in this comparison at all.

Microsoft 365 Copilot lives inside Word, Excel, PowerPoint, Outlook and Teams. Microsoft’s UK page lists it at £13.80 per user per month paid yearly, discounted from a £16.10 list price, and you need a Microsoft 365 Business plan underneath it before you can buy it. This is the one that drafts your reply and summarises the meeting you missed.

GitHub Copilot is a developer tool from Microsoft’s other house. Pro is $10 a month. It works inside code editors and, since February 2026, from a command line agent.

Claude Code is Anthropic’s agent. It comes with a Claude Pro subscription at $17 a month on the annual plan, and it is even included on the free tier in a limited form.

Here is the blunt version. If what you want is your inbox triaged and a Word document drafted from your own files, buy the Microsoft one and stop reading comparisons of the other two. If what you want is a job done to a folder of files, over and over, without you, keep going.

What is the difference between Claude Code and GitHub Copilot?

One is built around a code repository. The other is built around a folder. Everything else in this comparison follows from that single fact.

GitHub Copilot began as autocomplete for programmers and grew an agent on top. Its cloud agent takes a task the way a colleague would, through a GitHub issue or a pull request, works in the background, and hands back one pull request when it is finished. GitHub’s own documentation is clear about the fences around it. It only works with repositories hosted on GitHub. It cannot touch two repositories in one task. It works on a single branch. And it stops dead at 59 minutes.

If most of that paragraph read as a foreign language, you have your answer to the comparison already.

Claude Code makes a much simpler assumption. You have a folder. It reads what is in it, proposes what it intends to do, and does it once you say yes. There is no repository, no branch and no pull request, because there is no software project. The unit of work is a file, which happens to be the unit your business already runs on.

Now the fair part, because Copilot has closed some of this gap. Copilot CLI went generally available on 25 February 2026, and it genuinely edits local files and runs shell commands on your machine. It reaches outside tools through MCP, the same connector standard Claude Code uses. General availability also brought a mode that plans before it acts, an autopilot setting that runs a job end to end, memory that survives between sessions, and background delegation so a long task carries on while you get on with something else. It is a real agent, not a toy.

And one thing worth knowing before anyone tries to win this argument on model quality: Copilot CLI lets you switch between Anthropic, OpenAI and Google models in the middle of a session, Claude Opus 4.6 and Claude Sonnet 4.6 among them. You can run Anthropic’s models inside GitHub’s product. So this was never a question of whose AI is cleverer. It is a question of which product is shaped like your work.

Which one can work on a folder of invoices?

Both can reach your files. Only one behaves as though that was the point.

Picture the actual job. Forty PDF invoices in a Downloads folder, six suppliers, six different layouts, and you need supplier, date, net, VAT and total in a spreadsheet your bookkeeper will accept. With Claude Code you copy the folder, point at the copy, describe the columns you want and read the plan it comes back with before anything is written. With Copilot CLI you can do the same thing, but you arrive there through a tool whose entire surrounding product, its issues, its pull requests, its code review agent, assumes you are shipping software.

There is also a difference in how much rope each hands you on day one. GitHub’s documentation puts it plainly: Copilot has the same access to files on your computer as you do, and can run any shell command that you can run. Claude Code starts read only, asks before it edits anything, and stays inside the folder you launched it in unless you widen that yourself.

For a first attempt at automating something real, that default matters more than any benchmark score. If you want a quick read on which job in your business is worth pointing either tool at, the AI Opportunity Scorecard takes about two minutes.

What do they cost, and what changed in June 2026?

Claude Code is a flat subscription. GitHub Copilot stopped being one on 1 June 2026, and most comparison articles have not caught up.

Every Copilot plan now runs on GitHub AI Credits. Code completions and Next Edit suggestions are still included and do not consume credits, so if all you want is autocomplete, nothing has changed for you. Agent work is a different story. GitHub’s own announcement is candid about what went away: users who ran out used to fall back to a cheaper model and keep working, and that fallback is gone. Usage is now governed by available credits and administrator budget controls.

The allowances themselves are worth reading twice, because GitHub describes them two ways on two pages. The billing announcement talks in money, saying Pro includes $10 in monthly credits. The plans documentation talks in counts, giving Pro 1,500 credits made up of 1,000 base and 500 flex, Pro+ 7,000, Business 1,900 per user and Enterprise 3,900 per user. I cannot reconcile those units from the public pages, so treat any precise “how many tasks do I get” claim you read elsewhere as an estimate. The promotional top up that gave Business and Enterprise customers extra credits covered June, July and August only.

Anthropic’s side is easier to explain because there is less of it. Claude Code is included on Free in a limited form, on Pro at $17 a month annually or $20 monthly, on Max from $100, and on Team at $20 per seat annually. There are usage limits on all of them and Anthropic does not publish the numbers either, so neither company is being fully transparent here. The practical difference is what happens when you hit the wall. On Claude you wait. On Copilot the budget is spent.

Can you set either one up without a developer?

Yes for Claude Code, in about a minute. Probably not for Copilot CLI, unless you enjoy that sort of thing.

Claude Code takes about a minute.

1. Buy Claude Pro on the monthly plan, so cancelling later costs you nothing.
2. Open claude.ai/code in your browser. Nothing to install.
3. Drag in a copy of the files you want to work on, never the originals.

If you later want the job running on a schedule against files on your own machine, the desktop app for Mac and Windows handles that.

Copilot CLI asks for more up front. You need a GitHub account and a paid Copilot plan before an install even begins: Node.js 22 or later if you go the npm route, or Homebrew or WinGet instead, plus PowerShell 6 or newer if you are on Windows. Note the plan requirement, because GitHub is inconsistent about it. The changelog announcing general availability says the CLI is available with Pro, Pro+, Business and Enterprise, which leaves the free tier out, while the plans page states that all plans include Copilot CLI. Budget for the $10.

None of that is difficult for a developer. It is four unfamiliar decisions before you have automated anything, at exactly the moment your motivation is highest and your patience is shortest.

So which should you buy?

Claude Code, if you run a business and do not maintain software. I would not hedge on that.

The reason is not that Anthropic’s models are better, because you can run those same models through Copilot. It is that Claude Code asks you to understand one idea, a folder, while GitHub Copilot asks you to understand repositories, branches, pull requests and credits before it will do anything more interesting than autocomplete. Every one of those concepts is a place a busy owner quietly gives up on a Sunday afternoon.

There is a second reason, and it is about who you become. The skill you build with Claude Code is describing a job precisely enough that something capable can carry it out, then checking the result properly. That transfers straight into the briefs you write for staff and for every supplier you ever hire. It is the whole argument for learning this rather than renting it forever. If you would rather have someone find the highest value job in your business and build the first one live with you, that is exactly what the AI Opportunity Audit is for.

When GitHub Copilot is the better answer

When code is genuinely part of your business. Then Copilot stops being the awkward choice and becomes the obvious one.

If you employ or contract developers, your work already lives in GitHub and Copilot meets it where it is. The cloud agent picking up an issue, doing the work in the background and opening a pull request for a human to review is a real workflow, not a demo. Nothing in Claude Code replaces that review loop.

Copilot is also the right call if you already pay for Business seats across a team, or if what you actually want is faster typing rather than delegated work. On inline completion, the thing that appears as you type, Copilot remains the better tool and it does not spend a credit to do it. And if you have a technical person in the building who will do the setup, the $10 tier is a cheap way to get an agent with model switching built in.

The honest summary is that these two tools barely compete. They are aimed at different people and the search results pretending otherwise were written for the other one.

What I would do with your first twenty pounds

If I were in your position I would spend one month of Claude Pro, pick the single most tedious file based job in the business, and give it a Sunday. Not the most complicated job, the most tedious one. Copy the files first, never work on the originals, and make it explain its plan before it changes anything.

What I would not do is buy three subscriptions to compare them properly. That is how a decision that should take an evening turns into a project, and the invoices still are not typed up.

And if it turns out the thing you actually wanted all along was your email drafted and your spreadsheets tidied inside Office, go and buy Microsoft 365 Copilot with a clear conscience. It is a good product at a fair price and it is not what these two are for. If you want a second opinion on which of your jobs is worth automating before you spend anything at all, that is what I do, and there is more about how I work on my about page.