AI Without the Hype
You've sat through the keynotes. Strip out the breathless part and ask the only question that matters: what's actually useful on Monday?
For an experienced professional, treat AI as a fast, literal-minded assistant, not a junior partner. Right now its value sits in four specific places: drafting from your notes, summarizing piles of text, structuring messy information, and stress-testing your own arguments. Your job is to know where those four save you real hours without handing the tool anything you wouldn't trust a brand-new intern with.
I came to this late and skeptically, which I still think is the right posture. I ignored AI for almost two years because the marketing insulted my intelligence, all that breathless "transformation" language from people who'd never billed a client or carried a quota. I was wrong to wait that long; I left money on the table. But I wasn't wrong to distrust the pitch. The pitch is still mostly noise. The tool underneath it is not.
Why does so much AI advice feel useless to someone with experience?
Because most of it is written for someone trying to prove they're busy, not someone already accountable for outcomes. If you own a P&L, a client book, or a team's reputation, "automate your workflow" means nothing. Your work isn't throughput. It's judgment, relationships, and making the hard call when it's on the line.
You don't need to "10x your productivity." You need to stop burning your most valuable hours on work where your experience adds almost nothing: turning raw notes into a formal memo, taming an email swamp into a clean brief, reading 60 pages of dense material when you need the three sentences that matter. That's the whole reframe. AI doesn't upgrade your judgment; it takes the grunt work off your desk.
What can AI actually do reliably right now?
Narrower than the sales deck claims, wider than the skeptics admit. After enough real use with a tool like Claude, here's where it earns its keep, and where it predictably burns you.
- Drafting from your raw material. Give it your notes, bullets, or a brain-dump; it returns a structured first draft. That first pass has always been the slowest, most resistant part of knowledge work. Two hours of "getting started" becomes 20 minutes.
- Summarizing and extracting. A 60-page deposition, a dense contract, three years of board minutes: Claude pulls the structure and the key points in minutes. You still verify. But you start from a map instead of a blank page and a pot of coffee.
- Being a sparring partner. The underrated one. "Here's my argument, attack it. What's the best case against me?" At 11 p.m. it's a cheap way to lose the argument privately so you can win it in the room. I've caught more holes in my own logic this way than I'm comfortable admitting.
- Translating between registers. Turn a technical analysis into a board-ready summary, or a rambling email into three clean bullets. It's a register-shifting machine.
And where it fails, so you don't get burned: it invents facts and citations with total confidence, it has zero sense of consequences (it doesn't know which mistake merely annoys a client and which triggers a lawsuit), and it'll produce fluent nonsense in a field where you'd catch the error and a junior wouldn't. Your domain judgment is the safety mechanism. Which is exactly why this is a worse tool in a beginner's hands than a veteran's.
What is the "would I hand this to a sharp intern?" test?
One decision rule survives contact with real work. Before you give AI a task, ask: would I hand this exact task to a bright, fast intern who has zero context on my client and whose work I'd always review before it leaves the building? If yes, it's usually safe for an AI assistant, first drafts, summaries, research scaffolding, cleaning up language. Turning a two-hour call transcript into a one-page summary of decisions and open items is a perfect fit; the intern does the first pass, you do the final check.
If the answer is no, because being wrong is expensive, politically sensitive, or legally risky, then AI stays a research assistant only. You don't send, sign, or approve anything that hasn't gone through your own head first. This isn't ego. It keeps the hierarchy that works: the intern drafts, you decide, you sign. AI is just the fastest intern you'll ever have.
A composite from practice
Take a pattern I see often. Call him a 54-year-old independent CPA who'd written off AI as "for the kids" until a brutal tax season had him working most Sundays. He picked one grind of a task: the first draft of client advisory memos, the ones explaining a tax position in plain English. He'd feed Claude the facts and his recommendation, get a clean draft back, then do what he's always done, apply his judgment, catch the nuance, make it his. The memos went from about 90 minutes to 25. He didn't transform his firm. He bought back roughly six hours a week in his busiest stretch and spent them on the advisory conversations that actually grow a practice. No revolution. Just a mid-career professional refusing to do by hand what a tool now does adequately.
Hype claim versus what's actually true
| The hype says | What's actually true in 2026 |
|---|---|
| AI replaces professionals | It replaces the grunt work around the professional's judgment |
| It "thinks" and decides | It drafts and structures; you decide and you're accountable |
| You need to 10x your output | You need to reclaim hours, not multiply volume |
| Trust the output, it's smart | Verify everything factual; it invents with confidence |
| Younger people have the edge | Your domain judgment is the safety layer they lack |
| You're too late | The tools matured in 2025; you're early to the useful part |
How should I actually start without wasting time?
Avoid "AI projects." Pick one recurring, low-judgment task you personally find tedious and run only that through Claude for two weeks. Good candidates are text-heavy, dreaded, and don't carry final sign-off, meeting summaries, first-pass client emails, slide outlines, agenda docs. On a literal Monday morning, you could start like this:
- For a memo you're putting off: "Here are my rough notes on [topic]. Turn this into a clear two-page memo for [audience], with headings, a short intro, and three options with pros and cons. Leave anything you're unsure about in brackets for me to fill in."
- For messy meeting notes: "Here's a raw transcript. Give me a one-page summary: decisions made, open questions, and who owns what. Flag anything ambiguous instead of guessing."
- For a draft you've already written: "Here's my argument. Make the strongest case against it that a skeptical client would make, then tell me which of my points is weakest."
Verify everything. Keep your hands on the wheel. That's the whole strategy, and it's deliberately anticlimactic, the hype wants you feeling late and overwhelmed so you'll buy the course. The calmer reality: you have decades of judgment that make these tools more useful for you than for the people selling them. Use that judgment to find the line between what AI does well and what only you can do, and then stop doing, by hand, the work on the wrong side of it.