Why I've chosen Microsoft Copilot over ChatGPT and Claude Business

Microsoft’s latest Copilot announcement is not just another product update. It is Microsoft attempting to establish home pitch dominance over Microsoft 365 based businesses. Copilot was complex to set up and awkward to get started with, so many organisations set up ChatGPT and then Claude Business accounts and put them behind SSO. Even though Copilot was clearly the probable long-term winner, it was too painful to get value from and many wanted value fast.

It was clear from Microsoft Build in June that Microsoft’s approach is to commoditise models and focus on the application layer, and this is the logical next step.

What Microsoft announced last week

The new Copilot is being built around three ideas: Home, Code and Autopilot.

Home brings Copilot Chat, Cowork and Office together as a starting point for work.
Code lets people build small apps, dashboards, automations and workflows using natural language, powered by the same underlying technology as GitHub Copilot.
Autopilot is Microsoft’s persistent agent concept: a digital coworker with its own identity and permissions that can keep working when you are not sat watching it.

There is also a quieter but important commercial point: Microsoft is introducing FinOps for AI, because managing AI consumption is critical in the usage-based billing world we are increasingly in.

Why this matters for Microsoft 365 businesses

For businesses that bought separate ChatGPT or Claude accounts the win was the fast time to value. The pain was the difficult integration with the complexity of Microsoft 365. It never quite worked as well as it should. It always felt very clearly ‘separate’.

That pain is now the pressure point that Microsoft is kneeling on. Microsoft has been iterating Copilot in Microsoft 365 both on the web and in local apps continuously, and the experience has improved significantly and is now deeply integrated across the platform. That integration is hard to copy. For example, lots of businesses put a lot of effort into call transcription and reasoning over those transcripts with AI, but Teams and Copilot handle it out of the box.

It is also much easier to set up than it was, with a dedicated section of the Microsoft 365 Admin Centre. The fact that one can access models from OpenAI, Anthropic and Microsoft, with additional model providers likely to follow, is a compelling cherry on top.

Copilot Code makes vibe coding for internal apps safer by adding boundaries

The most interesting part of the announcement may be Copilot Code. It brings the power of GitHub Copilot but packages it in a simpler experience that knowledge workers can adopt fast, while aiming to keep the output governed and managed with Microsoft Copilot Managed Runtime. It moves closer to making vibe coding for internal apps enterprise-friendly.

I have seen a growing number of clients experimenting with Replit, Lovable, Claude Code and ChatGPT Codex to make internal apps. Being able to bring more of that experimentation inside the Microsoft tenant trust boundary could be significant.

To be clear, this will not remove the need for developers. Customer facing and more complex apps are still going to be built with GitHub Copilot or any of the above-mentioned coding agents, but it will likely remove the need for quite a lot of fragile internal spreadsheet-plus-macro-plus-email workflows, which is probably a public service.

Autopilot is the bigger organisational shift

Code is the eye-catching feature, but Autopilot may prove to be the bigger organisational shift. 2026 started with personal agent hype around OpenClaw assistants, then Hermes, and now Grok Bots and Facebook Muse. Microsoft teased Scout at Build in June but is now shipping Autopilot as a digital coworker with its own identity, permissions, workspace and persistent computer. If Microsoft executes well, this starts to look less like a chat window you prod for answers and more like something you can delegate work to and then leave running. We may be talking about delegated digital labour inside the existing Microsoft 365 operating model.

It will be interesting to see how this develops. Learning what tasks need to be delegated to a human, what tasks might be delegated to an Autopilot and what tasks are a waste of tokens is likely to become an important productivity skill for knowledge workers.

The obvious use cases are the tedious ones: chasing inputs for a supplier review, monitoring a project channel, preparing meeting packs, keeping a workback plan honest, or following up with stakeholders who have gone quiet. None of that is glamorous. It is also exactly the sort of work that consumes days of management time across a business.

This is not magic. You still have work to do.

The uncomfortable bit is that Copilot’s power makes the basics more important, not less. Copilot will only be as useful, and as safe, as the Microsoft 365 environment it can see. If your SharePoint permissions are a mess, your document naming is chaotic, your Teams estate has grown like a weed and nobody knows who owns which process, Copilot will not fix that for you. It will just expose the mess faster.

That is why the unglamorous Microsoft 365 configuration work grows in importance:

Sort out identity and access. Entra ID groups, conditional access, privileged access and the employee lifecycle need to be boring and reliable.
Review information architecture. Make sure Teams and SharePoint are logically structured to ensure access to only what is required.
Pick outcomes, not use cases. Work out the outcome and effect you want to have and work backwards to see how the technology might help.
Setup Microsoft Purview properly. This will help you make sure your sensitive data stays protected.

For smaller companies, be deliberate. Don’t give everyone every capability and hope for transformation. Pick a few early adopter champions, give them one workflow to improve, then have them show colleagues what worked: the prompt that saved half an hour, the meeting pack that built itself, or the tracker that replaced a painful spreadsheet. Copilot still needs disciplined setup, but adoption spreads when people see someone like them using it to get real work done.

Why I’m cancelling my OpenAI and Claude business accounts

I am cancelling them to save money, reduce complexity and consolidate my AI setup without reducing capability. If Copilot can provide strong models, business context, tenant controls, agents and lightweight app building, the simpler setup is the better one.

The same applies to larger Microsoft 365 organisations: one governed AI layer should be easier to support, and often cheaper, than a patchwork of standalone tools.

Bottom line

For the last couple of years, using separate AI tools made sense. Pick the best answer, keep it reasonably safe and move quickly. For Microsoft 365 businesses, I think that has now shifted. Copilot brings the models, context, controls and workflows closer together, which means less duplication, less admin and fewer exceptions to manage.

My recommendation is blunt: invest in Copilot properly. Clean up the foundations, pick real workflows, then use the platform you already pay for. Lower cost, less complexity and, assuming Microsoft executes, at least the same capability in a better governed place.

If you’re trying to make sure your team are well setup to take advantage of the benefits of AI, we can help. Reach out below.👇