Hope and Alarm

Last week, Anthropic published the results of the largest qualitative AI study ever conducted. Using an AI interviewer, they collected open-ended responses from 81,000 users across 159 countries and 70 languages.

The three things people said they most want from AI: professional excellence, personal transformation, life management. The top concerns: unreliability, job displacement, loss of autonomy.

Most people weren't neatly sorted into optimists and pessimists. Respondents who were most enthusiastic about AI for emotional support were also three times more likely to fear becoming dependent on it. Same capability, competing feelings. That's probably the most accurate picture of where we are right now.

Paradigm shifts have always looked like this. People thought the printing press was going to destroy the church. People thought cars were going to destroy communities. Some of that was true. Most of it resolved differently than anyone expected. A number of things can be true simultaneously: AI won't solve all our problems, it will make most of us better at our jobs, and some jobs will go away.

The conversation about AI and jobs gets flattened too quickly. Either AI takes everything or the fear is overblown. Neither framing is useful.

Here's what's actually happening: the most routine entry-level work goes first. Call center operations, basic data processing, first-draft anything. The writing is on the wall for those roles. Middle management won't disappear, but it's going to transform in a way that most managers aren't thinking about yet.

A middle manager today is responsible for two things: supervising a team and allocating a budget that the team spends. In an AI-native organization, that same manager will supervise a mixed team of humans and agents, and allocate a budget that includes both headcount and something most organizations don't even have a word for yet. Token spend.

How many runs through an LLM is your team authorized per week? Which workflows justify the cost? Which recurring tasks make sense to hand to an agent? Most organizations can't answer those questions because they haven't had to ask them. They will soon. This is a capital allocation problem and a human resources problem at the same time, and it lands on every manager's desk eventually.

If you haven't set up a Claude Project yet, last week's issue is the place to start.

A few small business owners have reached out recently asking about setting up AI agents to run around the clock. I think that’s the right instinct, even if most business owners are still trying to understand the basics.

Quick backstory on why this is relevant. A developer in Vienna built an open-source AI agent called OpenClaw in November 2025 as a side project. By February it had 145,000 GitHub stars and its creator had been hired by OpenAI. Within weeks, Perplexity shipped a competing product and Anthropic doubled down on Cowork. The race to build agents that can complete workflows, open apps, and send emails went from hobbyist experiment to mainstream product category in about ninety days.

I've been spending more time in both Claude Cowork and Claude Code lately. Cowork gives Claude access to a folder on your machine with full read and write permissions, and the ability to automate or schedule any recurring manual task. Code is for building software. Both are real, both are useful, and both are a few steps ahead of where most AI users are right now.

For the average AI user, the biggest near-term gains are still in Claude Chat. It can build a mini app, generate a mockup of a digital product, walk you through a complex decision. Get the reps in chat first, and once you start thinking in workflows rather than one-off prompts, Cowork starts making a lot more sense.

Read through the Anthropic survey mentioned earlier. There's an interactive quote wall filterable by region, vision, and concern. Spend ten minutes browsing, and keep an open mind. Important reminder that the entire world is grappling with AI, and it’s helpful to zoom out every now and then.

Negative sentiment toward AI is increasingly a luxury belief. The chart below, shared by Theo Jaffee of a16z, pulls from the Anthropic survey data. Respondents in Sub-Saharan Africa, Central Asia, and South Asia were the least concerned and the most optimistic. Respondents in Western Europe and North America were the most skeptical. The data on who's worried, and who isn't, should give you pause.

Someone built a real-time TSA wait time tracker using AI, and given everything going on right now, it couldn't be more timely. The app, built by Zach Griff, shows live wait times by checkpoint including PreCheck and Clear. TSA is everywhere in the news right now: staffing chaos, air traffic control issues, and a runway collision at LaGuardia this morning that killed both pilots. Someone saw a gap, used AI to close it, and shipped something useful.