AI in the patent industry: Instant coffee Copilot

It is now abundantly clear that AI cannot be ignored by the patent industry. However, amongst the private practice firms, there still remains a divergence of approach, with many attorneys still not using AI in their day-to-day practice. By contrast, every large in-house department is now using AI in some capacity. A large reason for this is Copilot, the “instant coffee” of AI.

Convenience beats quality

Few people actually like instant coffee or would choose it as their first choice of caffeinated beverage. You drink instant coffee at work because it is easy, quick and it gives you the caffeine you need. Copilot is similar. Copilot arrives bundled with the Microsoft 365 licence. Importantly, from a confidentiality and security perspective, it sits under the same enterprise agreement as everything else the firm already does in Microsoft (you should still check what it is connected to though, IPKat). It is therefore not surprising that in-house departments, sitting inside large corporates that run on Microsoft, have adopted Copilot wholesale as their AI of first-choice. 

This Kat has previously been rather dismissive of Copilot (IPKat) and, at the time, it deserved it. Early Copilot was remarkably poor compared to some of the other frontier LLMs, partly because of the interface Microsoft placed in the way and partly because of the models running underneath. Worse, because Copilot was the only AI tool people had access to at work, its rubbishness gave AI a bad name. 

This has changed in recent months. Copilot is now getting better, primarily because the best models from OpenAI and Anthropic are now available as options within it, but also because Microsoft's own models are improving. Copilot is still a long way from being the best AI tool available, but is now starting to be good enough. We are, incidentally, starting to see a plateau of AI model capabilities. They already have superhuman abilities, and there is a point where most of them are able to do what we need them to do in IP.

AI now works in Microsoft Word

Copilot is part of an important shift in AI over recent months that has kickstarted more widespread adoption in the IP industry. AI has moved out of the browser-based chat window and onto the desktop. The models can now see the context in your folders, open your files and work inside the applications in which you work. You can use them in our old favourites of Word, Excel, PowerPoint and Outlook. Importantly, they have also got better at this and at processing large project folders of context.

For a profession that lives in Word, SharePoint and Outlook, this shift matters a lot. You can now ask an AI to amend a claim set in tracked changes, in the document, without having to launch an entirely different system. There were some amusing predictions six months or so ago that the rise of AI for office work would finally instigate a shift from Word to a better format, such as markdown or LaTeX. As someone who wrote their thesis in LaTeX, this Kat had some sympathy with this as an ideal, but no expectation that it would ever happen. Patent attorneys are not moving out of Word and learning a new text editor, no matter how much better this would be. The biggest barrier to most patent attorneys using AI was that they couldn’t use it in Word or in their current folder systems.

Of course, Copilot is not alone in allowing attorneys to work within the familiar and comforting Microsoft environment. All of the big players now have their own desktop-based systems. Anthropic has a desktop application for Claude that can see your folders and work in your files, and Google offers Gemini throughout its enterprise Workspace, with full access to a firm's Google Drive. However, Copilot remains the leader because it is a simple add-on to the agreements that companies already have with Microsoft (and no-one wants to have to use Google Docs...).

You don't need to be an expert AI prompt engineer to use AI effectively

Getting started with AI is now also far easier. The models are better at understanding what you want, the tools sit inside the applications you already use, and prompts and instructions can be shared across a team. Of course, the tools remain imperfect, and knowing how to use them properly is still an advantage. However, this is less about being an expert prompt engineer, and instead about knowing what tasks a model is good at, and which it is not, and knowing how to avoid hallucinations and prompt biasing (IPKat). The difference is that this knowledge is now something every attorney can and should easily pick up (IPKat). We don’t all need to be expert prompt engineers, we just need some AI experience, and the best way of learning this is just simply to use it.

Of course, there is still the opportunity to take things further and do more with the tools available, should you have access to something beyond Copilot. In the “instant coffee” analogy, in a world of instant coffee, there is always a place for the coffee obsessives, with their own coffee bean grinders, carefully sourced single-origin beans, and obsessive water temperature control. These people make great espressos. But instant coffee will always be more convenient if you want widespread adoption in a large company, and it may be good enough for your purposes.

Expensive coffee pods

So where does this all leave the AI software wrapper companies? This Kat has previously argued that IP firms did not need these companies, on the grounds that attorneys should be upskilling themselves rather than paying for convenience, and that the wrappers added little over the underlying model (IPKat). This argument has now moved on, because the specialist IP AI wrapper companies are no longer even adding the convenience. Compared to Copilot sitting inside Word, or to the larger legal tech players integrated into a firm's existing systems, a specialist wrapper is a new system that your attorneys have to learn and an interface you are stuck inside, and integration with your own document systems is harder rather than easier. They are adding nothing, in terms of knowledge, context or capability, that the average attorney cannot now provide for themselves with a skill and a model that lives on their desktop.

Companies that grow up in the gap before a technology fully matures rarely survive. The functionality gap that the AI wrappers filled, namely the interface, the document handling and the prompt engineering, has now largely been closed by Microsoft and the frontier labs. Indeed, each of the big labs now also provides a legal specialist product of its own (Microsoft’s Copilot for Legal being the most recent). 

Interestingly, a 2025 MIT survey of enterprise AI adoption, The GenAI Divide, found professionals quietly abandoning the specialist tools their organisations had bought for them. A corporate lawyer at a mid-sized firm whose organisation had invested $50,000 in a specialised contract analysis tool nonetheless defaulted to ChatGPT for her drafting work, explaining that “Our purchased AI tool provided rigid summaries with limited customization options. With ChatGPT, I can guide the conversation and iterate until I get exactly what I need”. 

According to the report, the professionals surveyed trusted the generic AI tool more, were already familiar with it, and thought the answers were better. The same lawyer went on to say that “The fundamental quality difference is noticeable, ChatGPT consistently produces better outputs, even though our vendor claims to use the same underlying technology”. The report notably found that general-purpose tools often outperformed bespoke enterprise systems “costing orders of magnitude more”, at least on immediate usability and user satisfaction. This means that even if you spend a lot of time procuring an IP-specialist wrapper tool, you may still be at risk of AI shadow use and low adoption (IPKat). 

In the analogy, the wrapper companies are the coffee pods. They are expensive and not always obvious to use, they lock you into a single specialised machine for which the manufacturer can raise the price of the pods whenever it likes, and they come in lots of different colours that make no difference to what ends up in the cup. You are paying for the pod wrapper, and the simulacrum of convenience and quality. Those people who don't want the hassle of having to learn to use the pod machine may stick with their instant coffee. People who like good coffee will likely turn their noses up at the coffee quality of the pods and make their own caffeine arrangements. 

Final thoughts

The question is no longer whether patent attorneys will use AI. In-house, they already are, by and large through Copilot. Private practice will follow if they have not already done so, not least because clients who use Copilot every day will expect their attorneys to be at least as efficient. However, Copilot is the instant coffee of the AI world. It is not the best brew, but it has the benefit of being good enough for what is needed and is, most importantly, convenient with low barrier to entry. It is probably also a far more sensible choice for a large company than a coffee pod machine. Meanwhile, however, some people may still think it worth grinding their own coffee beans.

Further reading

Use of AI in the patent industry series:

• Use of large language models in the patent industry: A risk to patent quality? (October 2023)
• “Using AI tools to help assess inventive step”: A response to the CIPA journal article (June 2024)
• Solving the confidentiality problem (October 2025)
• The spectre of hallucination (October 2025)
• Are you behind the wheel or waiting for the bus? (February 2026)
• The future of the patent profession: Are we looking into an AI abyss? (March 2026)
• Is AI software for IP just expensive wrapping paper? (March 2026)
• Don't believe the hype. Believe the data. (April 2026)
• The risks of AI shadow use (May 2026)
• Bubble trouble and the affordability myth (June 2026)
• Beware the MCP Trojan horse (August 2026)
• The sneaky practice of prompt injection (and what to do about it) (October 2026)