Why I Ignored AI for Two Years and What Changed

May 28, 2026 • Anthony Guerriero

For two years, every AI headline was about something that sounded impressive in a demo and fell apart in practice.

AI-generated legal briefs that cited cases that didn't exist. Customer service chatbots that couldn't understand a basic question without escalating. Image generators that produced hands with seven fingers. Auto-drafted emails that somehow managed to sound both robotic and unprofessional at the same time. Every few weeks there was a new announcement about how this technology was going to transform everything, followed shortly by a quieter story about the ways it had done no such thing.

If you watched all of that and decided to wait, that was not a failure of imagination. That was sensible.

The legitimate reasons to wait

Two years ago, the tools were genuinely not ready for serious professional work. Not in the way that matters: consistent enough to trust, specific enough to be useful, careful enough not to confidently produce wrong information.

The accuracy problem was real. Early versions of these models would answer questions with a certainty that bore no relationship to whether they were right. A consultant asking about industry dynamics might get a plausible-sounding summary that got the numbers wrong and cited a trend that was already reversing. A lawyer asking about case law might get a citation to a ruling that never happened. The problem was not that the tools made mistakes — all tools make mistakes — it was that they made mistakes with confidence. That is a specific kind of dangerous for professional work.

The privacy picture was also murky. Who had access to what you typed? Was it being used to train other models? Were you inadvertently giving a technology company a copy of your client's most sensitive financial details? Those were legitimate questions in 2023, and the answers were not clear.

And the hype was exhausting in a way that made the tools seem worse than they were. When every product announcement is pitched as a revolution, the word loses meaning. It becomes noise. Reasonable professionals trained to filter noise did what they always do: they waited for the signal to separate from the static.

Waiting was not being behind. It was being careful.

What changed

Two things changed, roughly in order.

The tools got meaningfully better. Not incrementally — there were step-changes in capability, particularly in the ability to handle long documents, follow complex instructions, and maintain accuracy on professional-grade tasks. The hallucination problem did not disappear, but it became more manageable and more predictable: you learn where the model is reliable and where you need to verify.

More importantly, the evidence base shifted. In 2023, most of what you heard about AI in professional work came from tech journalists and startup founders. By mid-2024 and into 2025, you started hearing from working professionals — partners at law firms, senior accountants, management consultants, financial advisors — describing specific tasks where they were getting real value. Not theoretical value. Not "I could imagine this being useful someday." Specific: I used it to process 200 pages of discovery documents in three hours instead of two days. I used it to draft the quarterly client letter and it took 20 minutes instead of two and a half hours. I used it to prepare for a board presentation and it found three things in the financials I had not flagged.

When enough people you respect are describing specific results, the evidence base is different from when it is mostly demos and press releases.

The third thing that happened was visibility — the gap between the professionals who were using it and the ones who weren't started to show up in real ways. Not everyone's experience, not everywhere. But enough: the colleague who was handling twice the client load. The competitor firm that had materially faster turnaround times. The advisor who was spending client time on strategy because the document work was handled.

That is not fear. It is just an honest accounting of what the evidence shows.

The first concrete thing that moved the needle

The turning point was not a dramatic demonstration. It was a 40-page partnership agreement.

There was a client call scheduled for the following morning. The agreement had arrived the night before. Reading a 40-page partnership agreement at 10pm is technically possible. Reading it closely enough to prepare for a meeting — to identify the issues, the ambiguities, the provisions the other side might push on — takes a kind of focused attention that is hard to sustain late in the day.

The document went into Claude. The request was specific: "I'm meeting with the client tomorrow morning about this partnership agreement. They're coming in as a minority partner. Identify any provisions that give the majority partner unusual control, any notice requirements that could trap my client if they want to exit, and any economic terms that are structured to benefit the majority partner in ways that aren't obvious."

What came back was not a replacement for reading the agreement. It was a map. The three provisions that needed attention were flagged. The exit clause with the 180-day notice requirement was highlighted. The waterfall calculation that looked like a 50/50 split but wasn't was identified.

The meeting went well. The preparation that would have been inadequate was solid. The time investment was 12 minutes, not two hours.

That is not impressive in a demo sense. It is impressive in a this-changes-my-actual-workday sense. That is the distinction that matters.

What I wished I had known sooner

You do not need to change how you work before you start using it. That is the thing that kept me at the edge for longer than necessary — a vague sense that I needed to "learn AI" as some kind of prerequisite. That I needed to take a course, or understand the technology, or develop a whole new set of skills before I could use it effectively.

That is not how it works. You start with a task you already have. You describe what you need in the same language you would use to brief an intelligent assistant. You read what it produces, note what's off, and say so. That is the whole workflow.

The first use does not need to be your most important task. It needs to be real — something you actually have to do — so you can evaluate the output against something you actually know. A task you were going to do anyway, now done with Claude, tells you more than any demonstration.

And you will correct it. That is expected. That is the process. Claude produces a strong draft, not a finished product. The professionals who get the most from it are the ones who treat it like a capable first pass, not like a verdict. The correction is fast. The starting-from-zero is what used to take the time.

Two years of watching was not wasted. It meant arriving with skepticism intact, with real questions about accuracy and privacy and practical usefulness. That is a better place to start than enthusiasm.

But waiting any longer would have been a different kind of mistake. Not because of fear of being left behind — that framing is almost always wrong. Because the tools are now good enough to use, and the professionals who use them well are accumulating an advantage that compounds.

You are not late. You are under-leveraged.

The Leverage Starter — $199 founding access. A structured first week using Claude on your actual professional work. [Start here.]

Where this goes next

Want the guided, build-it-this-week version of this? See The Leverage Starter — or Turn Experience Into Income with Claude if you want the broader path.

Related reading from The Briefing

Not sure which program fits where you are? take the 2-minute course-fit quiz, or browse the full TLY course catalog.