TLDR; In my client work, adoption is easier to evaluate when a team starts with one task rather than a list of twenty-one. The recurring uses I see are writing assistance, code and data tasks, and meeting follow-up. I use a two-week pilot, then compare time, quality and rework before adding a second use case. These are practitioner observations, not a representative adoption study.
I’ve helped several teams start using AI at work. Most of the advice online skips the part that matters: what sticks after the first week.
The listicles are everywhere: “21 ways to use AI at work”, “101 AI tools for productivity”, “14 things Googlers use AI for every day”. The use cases in them are real. But after watching teams try and mostly abandon AI tools across two years of client engagements, the problem was never a shortage of ideas.
None of those lists tell you what to do on Monday morning, how long it takes to get useful results, or which of the 101 things is worth your time.
The three things that actually stick:
- Writing assistance: drafts, tone adjustment, editing
- Code and data tasks: formulas, scripts, spreadsheet automation
- Meeting follow-up: transcription, summaries, action items
What the listicles get wrong
The “21 ways” format has a fatal flaw. It’s optimised for impressiveness, not for adoption.
When someone hands you a list of 21 things, you try two or three of them. The first one is awkward and takes longer than expected. The second kind of works but the output needs heavy editing. By the third, you’re wondering if it’s worth the effort. A week later, you’ve quietly stopped.
This isn’t a motivation problem. It’s a habit formation problem. You can’t build a habit while jumping between 21 different tasks.
The other issue is that most lists conflate “technically possible” with “actually useful in your context.” AI can summarise a long legal document, generate marketing copy, analyse customer feedback, write Python scripts, transcribe meetings, draft press releases, review contracts. The question isn’t what AI can do. It’s what AI does well enough, for your specific tasks, in your specific workflow, that you keep coming back to it.
That distinction took me a while to internalise. Now it’s the first thing I ask new clients to sit with. If you want the fuller framework for where AI belongs in a workflow and where it’s better to build something deterministic instead, I’ve laid that out separately.
The three things that actually stick
I’ve watched a lot of teams adopt AI tools, and watched them abandon those tools. After enough cycles, the same three uses are what I still see in place six months later.
Writing assistance
This is the use case that has persisted most consistently in my client sample. I have not maintained a formal denominator or controlled cohort, so I no longer attach a percentage to it. Across the teams I can follow beyond the initial engagement, some form of AI-assisted writing is the recurring pattern I see most often.
What that looks like in practice: using AI to draft a first version of something you’d otherwise stare at a blank screen for. An email response. A project brief. A job description. A proposal introduction. A meeting summary. The AI draft is rarely perfect, and it shouldn’t be. It exists to get something on the page you can edit, which is almost always faster than writing from scratch.
I also see a lot of people use AI for tone-matching: taking something they’ve written and asking for it to be made more formal, more concise, or more friendly. That one is underrated. I use it regularly myself.
What I don’t see stick: using AI to write things wholesale without review. The output sounds like AI, smooth, competent, a little soulless, and people who receive it often know. More on that below.
Code and data tasks
This one surprises non-technical people, but it’s real. You don’t have to be a developer to use AI for code and data work.
The pattern I see most often is people using AI with spreadsheets. They have a spreadsheet, they need to do something with it that they don’t quite know how to do. A formula they can’t remember. A way to restructure the data. A macro to automate something repetitive. Instead of spending 40 minutes searching Stack Overflow or calling their IT person, they describe the problem to an AI in plain English and get a working formula or a step-by-step walkthrough.
For people who write code, the time savings are larger. Not replacing the thinking, that still has to be yours, but cutting the gap between “I know what I need to build” and “I have a working version to iterate on” from an afternoon to a few minutes.
The bar for this to stick isn’t “you need to learn programming.” The bar is: can you describe your problem clearly? If yes, you can probably get useful help with code and data tasks.
Meeting follow-up
I was sceptical about this one early on. Transcription tools have been around for years and the quality was never quite good enough to trust. That’s changed.
Teams who adopt AI-assisted meeting follow-up tend to keep it going. It removes one of the most tedious tasks in office work. Turning a 45-minute meeting into a usable summary of decisions and next steps used to take 20-30 minutes of someone’s time. Now it takes two minutes of review.
The friction point is setting it up. It requires connecting a tool to your calendar and meetings, and some people never get past that step. Teams that do set it up almost never go back.
Update (July 2026): Zoom, Microsoft Teams, and Google Meet have since added native AI meeting notes (AI Companion, Teams intelligent recap with Teams Premium or Copilot, and Gemini “take notes for me”), available to teams on the right licence tier once an admin or meeting host turns them on. If your organisation already has one of these, start there before adding a separate tool. Otter still earns its keep for teams without that licence tier, or teams that want one tool across every platform.
How to start
The biggest mistake I see teams make is trying to adopt “AI” as a category, rather than adopting a specific tool for a specific task.
The advice I give every new client is boringly simple: pick one boring task you do every day, and use AI for it for two weeks straight. Don’t add another use case until week three. Don’t try a different tool. Don’t read about prompting techniques. Just do the one task.
I use two weeks because it is long enough to encounter repeated examples of the task without turning the pilot into a quarter-long programme. It is an operating window, not a universal habit-formation threshold. During that period the team encounters the real friction: awkward prompts, output that needs editing, and adjustments to how the task is described.
After two weeks, you’ll have one thing that works. Then you can add a second.
This sounds slow, but it produces evidence the team can use. In my engagements, narrow pilots have been easier to sustain than broad launches. I have not run a controlled comparison proving that the pilot design causes one-year retention.
Before you pick your task, it helps to know whether your organisation is in a position to get value from AI adoption, not every team is at the same starting point. If you want to be more systematic about that, read the AI readiness assessment I run with every new client.
What to watch out for
I’d rather tell you this upfront than let you learn it the hard way.
Generic AI text is noticeable. AI writing has a characteristic quality: thorough, well-structured, slightly impersonal, oddly confident about everything. Readers who see a lot of written communication are starting to recognise it, especially in contexts where they expect a personal voice. If you’re sending a client email, a LinkedIn post, or anything where your personality matters, treat the AI draft as a starting point and put yourself back into it. Edit until it sounds like you. If it doesn’t sound like you, it isn’t done.
Be careful with confidential information. Consumer AI services normally process inputs on provider-controlled infrastructure. Retention, model-training use, regional processing, administrative controls, and contractual commitments vary by product, plan, settings, and organisation agreement. Do not assume a consumer account is approved for client data, financial information, legal matters, or personally identifiable information. Check the provider terms and your organisation’s policy, minimise what you send, and use an approved enterprise arrangement where the data requires it.
The “faster to do it myself” response is useful evidence. Early attempts can be slower because you are learning how to describe the task, when to trust the output, and how much editing it needs. Some friction falls with practice; some indicates the task or tool is a poor fit. That is why the two-week pilot ends with a comparison of time, quality and rework rather than an instruction to keep pushing regardless.
The tools I recommend to non-technical teams
Vague tool recommendations are useless, so let me be specific.
Claude (claude.ai) for writing tasks. This is my go-to for anything involving language: drafting, editing, summarising, rewriting, tone adjustment. The quality on writing tasks is consistently high in my work, and it handles nuanced instructions well. “Write this more concisely but keep the main argument” or “make this sound less formal without sounding casual”, Claude handles these reliably. Anthropic lists Pro at USD$20 before local pricing and tax; check the current checkout rather than relying on a converted figure in an article.
ChatGPT for research and quick lookups. ChatGPT with web browsing is useful when you need current information: recent news, competitor pricing, what happened last week. I also use ChatGPT for quick one-off questions where I want a fast answer, not a carefully crafted piece of writing.
Otter.ai for meeting transcription and summaries. For teams who want to fix their meeting follow-up problem, first check whether your video-call platform already does this natively: Zoom, Teams, and Meet all ship AI notes now on the right licence tier (see the update above). Where that’s not available, Otter is my standard recommendation. It integrates with Zoom, Google Meet, and Teams, produces readable transcripts, and generates summaries with action items. The free tier handles a reasonable volume; the Business plan is worth it if your organisation runs a lot of meetings.
A spreadsheet AI for data tasks. If your team lives in Excel or Google Sheets, the AI features built into both platforms may cover formula help, restructuring and summaries. Availability depends on the licence, administrator settings, region and workbook or data permissions. Check what is enabled in your existing account before buying another product, but allow for configuration and governance work.
One note on tools: don’t spend too long evaluating. The tool matters less than the habit. Pick one from this list, use it for a specific task for two weeks, and form an opinion from experience rather than from reading more reviews.
Where this goes from here
Once you’ve built a few stable AI habits, the question changes from “how do I start” to “where does this take my organisation.” That’s a more interesting question, and a harder one.
Some teams do this gradually on their own. Others hit a ceiling: they have individuals using AI tools, but they haven’t changed how their workflows or processes are structured to take advantage of it. The individual wins are real but they aren’t compounding into something bigger.
That’s usually when the conversation shifts from “which tools should we use” to “how do we design our work around these capabilities.” It’s a different kind of engagement, and if that’s where you’re heading, my Applied AI work is what covers it.
If you want a clearer picture of where your organisation sits before you get there, the readiness assessment I run is the place to start, now as part of that Applied AI work. It’s a short conversation, not a 75-question survey, and it gives you a realistic picture of where the blockers are, not just a score.
The tools are accessible and the information is everywhere. Making it stick is the hard part, and in my client work it is the difference between the teams still using AI six months on and the ones who tried it at an offsite.
Start with one task. Use it for two weeks. See what you notice.
Sources for current product capabilities
- Claude plans, Anthropic. Current plan structure and US pricing. Primary documentation
- ChatGPT Search, OpenAI. Current-information search with linked web sources. Primary product announcement
- Meeting Summary with AI Companion, Zoom. Meeting-summary capability and account requirements. Primary documentation
- Recap in Microsoft Teams, Microsoft. AI recap features and licence requirements. Primary documentation
- Take notes for me in Google Meet, Google. Meeting-note features, eligible plans and administrator controls. Primary documentation
Common questions
How can I use AI in my work?
In the client teams I have worked with, the recurring uses I see most often after six months are writing assistance, code and data tasks, and meeting follow-up. This is a practitioner observation rather than a representative adoption study. Pick one relevant task, measure whether it helps, and build from the result.
How do I start using AI at work as a beginner?
Pick one boring task you do regularly and test AI on it for two weeks without adding another use case or switching tools. Two weeks is the pilot window I use with clients, not a scientifically established habit threshold. At the end, compare time, quality and rework before deciding whether to continue.
How can I use AI for office work?
The most durable office use cases are writing assistance (drafting emails, project briefs, tone adjustment), spreadsheet and data tasks (formulas, restructuring data, automating repetitive steps), and meeting follow-up (transcription and action-item summaries). I recommend Claude for writing tasks, a spreadsheet AI like Copilot or Gemini in Sheets for data tasks, and for meeting transcription, check whether Zoom, Teams or Meet already includes AI notes on your licence tier before adding a separate tool like Otter.ai.