Morning guys, happy Saturday.
I’ve been thinking about how often we blame people for decisions that were mostly shaped before they got to make them.
A user chooses the wrong mode, but we gave them six modes. A patient misses an appointment, but the train was late and missing work would cost them their job. A founder keeps chasing the wrong things, but their environment is feeding them the same information every day. An investor makes a bad call, then explains it as an isolated mistake even though they have made some version of the same mistake three times.
We tend to focus on the visible decision because that is the part we can see. I think the more interesting stuff often happens one layer earlier, in the interface, environment, incentives and habits that made one choice much easier than another.
There were five good examples of that this week.
Sponsored by…
The fastest teams stopped typing.
Your day is Slack threads, emails, docs, and CRM updates. Every one of them starts at a keyboard, and the keyboard is the bottleneck.
Wispr Flow turns your voice into clean, ready-to-send text in any app. Speak a three-paragraph client update in 30 seconds. Dictate the proposal while it’s still fresh. Flowstrips filler, fixes grammar, and formats it so you just hit send.
4x faster than typing. Used by teams at 270+ Fortune 500 companies. Works on Mac, Windows, iPhone, and Android, with Flow for Teams for your org.
Try Wispr Flow Free →
1. Anthropic finally stopped making users understand Anthropic
Anthropic merged Claude Chat and Cowork into one interface last week. Cowork is not actually gone, despite some of the headlines. The capability still exists alongside Artifacts, Design and the new Docs and Slides features. What disappeared is the need to decide, before doing any work, which Anthropic product your work belongs inside. Claude can increasingly make that choice itself.
This sounds like a fairly boring UX change, but I think it points toward where AI software is going. Traditional software makes us translate what we want into the structure of the product. If I want to make a presentation, I first choose Slides. If I want to analyze something, maybe I open a spreadsheet. If I want to design something, I open Figma. We have spent years learning which tool corresponds to which type of intent because the software itself was not smart enough to make that connection.
Once the software understands the request, though, the whole thing starts to look backwards. Why should I know whether my request needs Cowork, Docs, Design or some internal Anthropic capability? I just know what I am trying to get done. Asking me to select the implementation is a little like asking someone ordering dinner which pan the chef should use.
This problem shows up everywhere because products tend to inherit the structure of the company that built them. You have a team responsible for analytics, another for automation and another for reporting, and eventually the customer gets three tabs called Analytics, Automation and Reporting. Internally, that makes perfect sense. Externally, you have handed the customer a small version of your org chart and asked them to learn it.
AI gives companies a chance to reverse that. Products can become much more complicated underneath while becoming simpler on top. I suspect this will be one of the better tests for whether something is genuinely AI-native. If every new capability produces another mode, tab and selector, you might just be putting AI inside the old software model.
The better product may be the one where the user says what they want and never has to care which product team built the answer.
2. Market research is getting cheaper. Real depth is not.
Rex Woodbury tells a story about asking Priscilla Chan how she thinks about depth versus breadth of impact. Chan talked about something she noticed while working at UCSF Medical Center: one of the things determining whether some patients actually received care was whether BART was running on time. If the train was delayed, the patient might skip the appointment rather than risk arriving late to work.
I like this example because I cannot imagine it appearing near the top of a normal healthcare market map. If you approached the problem from the outside, you would probably start with insurance, physician availability, wait times, hospital capacity and cost. All reasonable things. You could ask Claude to produce fifty pages on barriers to healthcare access and end up with a very intelligent document that never makes you care about one delayed train.
That is the difference between knowing more about a market and being inside it long enough to know what actually matters.
This matters more now because the first kind of knowledge is becoming ridiculously cheap. Anyone can generate a competitor matrix, summarize reviews, map an industry and produce a list of underserved niches before breakfast. That is useful, but it also means thousands of people can start with essentially the same map. The advantage moves toward whatever does not make it into the map.
Usually that is some small, annoying detail people inside the industry have stopped noticing. A spreadsheet everyone emails around because the official software is useless. A manual approval step everybody assumes has to exist. The thing customers complain about versus the thing that actually makes them leave. A train schedule that determines whether someone sees their doctor.
People talk about founder-market fit as though it means being passionate about an industry. I think the practical version is much more valuable: you have spent enough time there to rank the facts differently from everyone looking in from outside.
AI can help us understand what people already know. Depth is how you notice the thing nobody thought to write down.
3. The useful part of the 80/20 rule is choosing a side
Greg Isenberg has an 80/20 list for founders that keeps getting passed around: retention over acquisition, existing customers over new leads, listening over pitching, word of mouth over paid advertising, customer experience over branding and so on.
I would not take the numbers literally. There is no universal law saying that a company should spend exactly 80% of its effort on retention and 20% on acquisition, and this is not really the Pareto principle in any rigorous sense. But I think arguing over the percentages misses why lists like this are useful.
They force you to stop saying that everything matters.
Most business advice is annoyingly hard to disagree with. Product matters. Distribution matters. Customers matter. Brand matters. Growth matters. Quality matters. Everyone nods, then keeps doing everything they were already doing because nothing in that list forces a tradeoff.
The moment you say retention matters more than acquisition right now, something has to lose. If listening matters more than pitching, maybe you do not spend tomorrow writing another sales deck. If existing customers matter more than new leads, perhaps the founder joins three customer calls instead of opening another outbound channel. Strategy gets more useful once two good things are placed against each other and one of them wins.
There is also something interesting about which activities usually end up on the weaker side of these lists. They tend to be the visible ones. Getting a new logo feels like progress. Launching a campaign creates screenshots. Redesigning the website gives everybody something new to look at. Keeping an existing customer happy is much less theatrical. Nobody posts a celebratory Slack message because a customer quietly renewed for the fourth year.
Businesses naturally drift toward work that looks like work. I think a good 80/20 list is mostly a device for catching yourself doing that.
So I would not ask whether Greg’s percentages are correct. I would ask which pair makes you slightly uncomfortable because you know you have been spending most of your time on the easier side.
That is probably the useful one.
4. Sometimes you are not stuck. Your context is.
Zan Tafakari writes about something I keep noticing in my own work: we treat thoughts as though they come entirely from inside us, even though the environment around us is constantly deciding what gets into our heads in the first place.
This sounds obvious until you think about how knowledge work actually looks now. Same laptop, same desk, same Slack channels, same browser, same feeds, five days a week. Then you get stuck on a problem and the response is usually to stare at the same screen harder. Maybe open another tab. Maybe ask another model. Maybe make another coffee and sit back down in exactly the place where you have already failed to solve it for three hours.
I do this all the time, and I think there is a bad assumption underneath it. We assume the answer is already somewhere inside our heads and the only thing missing is more computation.
Sometimes the missing thing is new input.
Go sit with the person using the product. Take the call while walking. Visit the warehouse. Work somewhere else for an afternoon. Talk to someone who has nothing to do with your industry. Read something that has nothing to do with the problem. You are not guaranteed an epiphany, obviously, but at least you have changed the information entering the system.
There is a reason good field research can beat weeks of dashboard analysis. A dashboard shows you the variables someone already decided were worth measuring. Walking around the real thing gives you a chance to notice variables nobody put in the dashboard.
This is one area where AI might actually make us worse if we are not careful. Models make it possible to do more and more intellectual work without leaving the same interface. Research, brainstorming, writing, coding, analysis and planning can all happen in one chat window. That is incredibly useful, but it can also make the information environment narrower while making it feel broader.
You can have twenty conversations without encountering anything genuinely unexpected.
Sometimes the smartest thing you can do with a problem is stop giving your brain the same world.
5. You probably have recurring bugs too
Investor Chris Paik has a good essay about trying to outsmart yourself. His argument is that investing has an obvious outward-facing component, studying companies, markets and data, but also an inward-facing one that gets much less attention: understanding the person making the judgment.
That sounds soft compared with building a model or analyzing a market, which is probably why people avoid it. Markets can be researched from a comfortable distance. Looking at your own bad decisions tends to involve discovering that the problem was greed, ego, impatience, fear of missing out or some other explanation that is considerably less flattering than “the thesis changed.”
What I like about Paik’s approach is that he makes self-knowledge operational. He has noticed, for example, that doing his own work on an investment matters because his willingness to do that work is itself information. If he cannot get interested enough to properly investigate something, he should probably pay attention to that instead of finding a clever reason to invest anyway.
Software people would recognize the idea immediately if we changed the nouns. A program produces a bad result. You inspect the logs, reproduce the issue and eventually discover that what looked like five unrelated failures all came from the same bug.
People are very good at treating the equivalent failures in their own lives as unrelated events.
You hired too quickly because that candidate was unusually convincing. You invested too early because this market was different. You kept the product alive too long because customers really did love it. You abandoned another one too early because the market was obviously too small. Each story sounds reasonable on its own. Put ten years of them next to one another and you might discover the same person making the same decision for the same emotional reason.
This is why I like decision journals, even though the phrase makes them sound more serious than they need to be. Before an important decision, write down what you believe, what would prove you wrong and what is actually driving the decision. Then come back later and compare it with what happened.
The useful part is not predicting everything correctly. It is removing your ability to quietly rewrite history.
Maybe you repeatedly mistake urgency for importance. Maybe charismatic people make you abandon your skepticism. Maybe you keep projects alive because stopping feels like admitting you wasted time. Maybe every time somebody else starts succeeding in a category, you convince yourself you need to enter it too.
Those patterns are much more valuable to know than the details of any one mistake because they travel with you into the next decision.
We would never ship an important piece of software, watch it fail repeatedly and refuse to inspect the underlying code. Somehow we are much more comfortable doing that with ourselves.
If you’ve been enjoying Token Economy, subscribe and share it with someone who might like it too. Each week, we look at what’s changing in AI, tech, startups, and business, why it matters, and where things might be heading next.