The Optimists Are the Ones Holding the Tools

Hey Everyone - Hope you had a great weekend. A new poll landed last week that cuts against the picture you get from tech coverage, where everyone is using AI and the only debate is how fast it improves. The actual adoption data tells a different story, and the split inside it explains most of the optimism-versus-doom argument that fills my feed every day. Let me walk you through it.

This week:

Let's dive in.

If you read tech coverage, everyone is using AI. If you ask American workers, most of them are barely touching it.

A new Ipsos poll for the Groundwork Collaborative, fielded in June with 1,533 employed Americans on a probability panel, found that three in ten workers use AI at least weekly. Flip that number around. Seventy percent of the American workforce is using these tools less than weekly, three and a half years into the biggest technology rollout of the decade.

Who the users are matters more than how many there are. Roughly half of workers earning $100,000 or more, workers with bachelor's degrees, and white-collar employees use AI at least a few times a month. For workers earning under $50,000, workers with a high school education or less, and blue-collar employees, adoption drops to a quarter or less. The tools are concentrating exactly where income and credentials already concentrate.

Now the finding I keep turning over. Two-thirds of all workers expect AI to make the working experience worse, through eliminated jobs and increased pressure, rather than better through freed-up time. That pessimism holds across race, gender, education, and income. Everyone agrees on where this is heading for workers as a group.

Ask about their own job and the consensus breaks. Nearly half of college graduates believe AI will improve their job. Among workers with a high school education or less, one in five believe that. And the people with the darkest read are the ones already outside: unemployed and marginally attached respondents were the most likely to say AI is already hurting their line of work.

Sit with that structure for a second. Workers broadly expect AI to be bad for workers, while the credentialed half expects it to be good for them specifically. Both beliefs can be reasonable at the same time, and the whole optimism-versus-doom debate that fills your feed compresses into this one survey. The people writing confident posts about AI abundance are overwhelmingly drawn from the group that uses the tools daily, benefits from them today, and works in jobs where AI currently assists more than it replaces. Their optimism is real. It is also a report from one side of a widening split, and it gets mistaken for a forecast about everyone.

Here is why this matters for the way this newsletter thinks about hedging. The hedges I keep coming back to (local community, independent income, physical and relational skills, owned distribution) share a property that most AI-era career advice lacks. They are available on both sides of the split. "Learn to prompt better" is advice for the half that already has the tools. A reputation in your town, a trade someone will pay for directly, a list of people who trust you, a neighbor who vouches for you: those are open to a warehouse worker and a data scientist alike, and they hold their value in either future.

One more implication worth naming. If you are on the tool-rich side of this divide, your sense of how AI is going is drawn from an unrepresentative sample that includes you. The 70 percent are the majority, their pessimism is grounded in their actual position, and any picture of the AI economy that leaves them out is a picture of a minority experience wearing the word "everyone."

Part of my work is still consulting in the sports analytics space, and I want to be honest about something uncomfortable in it. The impact is close to unmeasurable. The metric that matters is wins, and wins have too many contributing factors (talent, health, coaching, schedule, luck) for anyone to isolate what our work added. We are fairly confident there is a net benefit. Proving it against the core metric is somewhere between hard and impossible, and that ambiguity hangs over every renewal conversation.

The newsletter agency sits at the other end of the spectrum. We have a set price. Working with us creates a new cash flow channel for the client, and that channel has a number on it that we are held accountable to. We are confident that most clients fully return their investment with us inside two to six months, and they can check our math on their own dashboard.

Selling the two could not feel more different. The clear offer with an expected payback period almost sells itself, because the buyer can picture the spreadsheet. The consulting requires faith in a chain of reasoning that ends in "trust us, this probably helps." Same me, same effort, wildly different conversations.

The agency structure is also a real win-win. I get paid, and our clients make more money than they would have without us. Everyone at the table can see the surplus, which makes the relationship easy and the renewals easier.

Both models work. But ease of understanding turns out to be a huge value add all by itself, separate from the underlying quality of the work, and I am now indexing on it. When I evaluate what to build next, the first question has become: will the customer be able to see the return without squinting? If the answer is no, the offer needs redesigning before it needs marketing.

Before you adopt AI for any meaningful task, write down what the task costs you today. That is the whole skill. Time per instance, error rate, how it feels, what it produces. Ten minutes of notes, captured before the tool arrives.

The reason this matters comes from watching the largest companies in the world skip it. A new Plug and Play survey covered in Forbes found that 74 percent of the biggest global companies now run AI in production, and roughly half cannot demonstrate business value. The diagnosis is procedural. They deployed without pre-established baselines, so attribution became impossible after the fact. They have spent billions and cannot tell you what changed, because they never wrote down what "before" looked like. KPMG's June survey of 2,145 leaders found 7 percent with established AI ROI.

This connects straight to the Building section above. The reason my agency offer is easy to sell is that the client can measure the return. The reason my consulting is hard to sell is that nobody measured the before. A baseline is what turns "I think this helps" into "this returned 4x," and the difference between those two sentences is pricing power.

It also protects you personally, and this half matters more. The deskilling research in this week's Learning section keeps finding the same shape: the visible metric improves while the invisible one erodes, and the people involved cannot feel it happening. Your unassisted baseline is the only honest reference point you will ever have. If you know a client proposal took you three hours and a 15 percent revision rate before AI, you can check in six months whether you got faster, whether quality held, and whether you can still do it at all without the tool. Skip the baseline and in six months you will have opinions and no evidence.

The practice, concretely: pick the three tasks where AI is most embedded in your work. For each one, write down what it costs you today, with the tool, and once without it. Date the note. Revisit quarterly. It is the cheapest data you will ever collect, and it only exists if you collect it now.