Deep and Wide: Deep Work After GenAI

I sat down for two hours and did good work. I built out a product-manager profile, fixed a pile of bugs, landed a couple of PRs. Real output. The kind of afternoon you end feeling like you actually moved things.

And trydeepwork couldn't see a second of it.

There was no deep work session for any of it, because I wasn't doing one thing. I was running maybe six threads at once — handing a chunk to an AI agent, jumping to another while it churned, coming back, reviewing, redirecting, shipping. The whole shape of the work was wrong for the tool I built. trydeepwork is a monument to doing one thing. I'd just spent two hours doing eight.

That gap bothered me enough to sit with it. This post is where that thinking landed, and why we're changing what the product measures.

The thing the tool was built to forbid

trydeepwork has a constitution — six tenets I try not to violate. The very first one is Forced Single-Tasking: a session is one task, the interface won't let you switch, and that constraint is the product. It's the thing you're paying for. Most of the value of a focus tool is that it says no on your behalf.

So my afternoon didn't just fail to fit. It violated the founding rule. My first instinct was the honest one: that wasn't deep work, it was glorified context-switching, and the tool is right to ignore it.

But that didn't sit right either. Because the work was hard. Holding six moving threads in your head, deciding what to hand off and what to keep, catching the agent's mistakes, keeping the architecture coherent across all of it — that is not a low-effort state. I was as cognitively taxed as I am writing a tricky function alone. It didn't feel shallow. It felt like deep work of a different kind.

That tension — "this breaks my most important rule, but it's clearly real work" — is the whole story.

Single-tasking was always a proxy

Here's the realization that cracked it open.

Go back to what "deep work" actually means. Cal Newport's definition is "professional activity performed in a state of distraction-free concentration that pushes your cognitive capabilities to their limit." Read it again and notice what is not in there: the number of tasks. Deep work was never defined as one thing. It was defined as full cognitive engagement.

So why did I build the whole product around single-tasking? Because, before AI, doing one thing was the only practical way to be fully engaged. Concentration and single-tasking were so tightly correlated that I treated them as the same fact. They were never the same fact. Single-tasking was a proxy — a cheap, reliable stand-in for the thing I actually cared about, which was engagement.

I confused the mechanism for the goal. The tenet protected single-tasking when it was really supposed to protect depth.

GenAI broke the proxy

For most of history, the proxy held, so the error never showed. AI is what pulled the two apart.

When you can offload real chunks of work to a capable agent, a new mode of working appears: you become the conductor. You hold a large context, you direct several streams at once, you operate at high tempo — and you can do all of it locked in, pushing your limit the entire time. It's the cognitive posture of a conductor running an orchestra, a pilot in a cockpit, a founder in a war room. Lots of threads, total engagement.

Once that mode exists, the single-task rule starts making two opposite mistakes at once:

• False negatives. It ignores my two hours of engaged orchestration entirely. Invisible.
• False positives. It hands full "deep work" credit to someone who picked one task and then spent ninety minutes half-watching Twitter — because they satisfied the rule.

A measure that's wrong at both ends isn't measuring the right thing. The proxy is broken. The honest move is to stop measuring the proxy and go back to measuring the goal: were you engaged?

Two shapes of focus: deep and wide

So we stopped thinking of it as "deep work versus multitasking" and started thinking of it as two shapes of the same thing:

• Deep Work — one hard problem, sustained, going as far down as you can on a single thing. The classic. The dive. The bunker.
• Wide Work — many live threads, a large context held at once, high tempo, full engagement. Orchestration. The span. The console.

Both push your cognitive limits. Both are real, engaged focus. They feel different, they suit different problems, and a serious tool for serious work should be able to see both. trydeepwork could only ever see the deep, single-threaded kind. That was the blind spot.

(A note on the names: the new mode is called Wide Work — it pairs with Deep Work and joins the microWork family. You're the operator at the console; we just don't make "operator" the label.)

The trap we almost walked into

The obvious next step is to ask: fine, but how does the tool know whether your wide session was engaged orchestration or just distracted thrashing through busywork?

The tempting answer is the rating. trydeepwork already asks you to rate each session 1–5, "very distracted" to "flow." Just gate it: rate it flow, it counts as deep; rate it distracted, it doesn't.

We almost did exactly that. It's wrong, and it's worth saying why, because the reason became a new principle for us.

The moment a self-reported number decides whether you get credit, it stops being a measurement and becomes a target. And targets get gamed. You'd start rating everything "flow" to earn the credit — not even dishonestly, just human. The rating, which today is a genuinely useful private signal about how your work feels, would turn to garbage. It's Goodhart's Law, and it's iron: when a measure becomes a target, it ceases to be a good measure. Worse, it makes the tool into something that judges you, when you opened it to help yourself.

So: instrument, don't judge

This is the part I'm proudest of, and it's bigger than this one feature.

The mistake underneath every version of this problem was assuming the product's job is to grade your work — to stamp "deep" or "not deep" on each session. Grading is what invites the gaming, and grading is what makes a tool feel like a critic. So we're taking grading off the table.

trydeepwork's job is to be an honest instrument, not a judge. A console of accurate dials. You're the operator; you read the instruments and decide what your day meant. The product never certifies your work as deep. It just shows you your activity clearly enough that you can tell for yourself. (If you've ever noticed the app reads like a terminal dashboard rather than a cheerful consumer app that congratulates you — this is why. It was always meant to be an instrument.)

That gives a clean rule for which signals can carry weight:

• Objective, behavioral signals can have consequences — time spent, how many tasks a session touched, what actually moved. You can't meaningfully fake sitting there and working for two hours, and "how many threads did this session touch" is just an observed fact, not a claim you have to justify. Deep versus wide falls right out of it: one task is deep, many is wide. We know the shape without asking you to rate or defend anything.
• Subjective, self-reported signals stay pure mirrors — the 1–5 rating remains exactly what it should be: a private note to yourself, for spotting your own patterns over time. It gates nothing. It scores nothing. Because the instant it did, it would be worthless.

So we don't decide whether your orchestration was "real" deep work. We capture it honestly — by time, by shape, by what moved — keep your felt sense of it private and consequence-free, and show you the readout. You look at it and conclude, yeah, those locked-in orchestration blocks were deep work. The tool gave you the evidence. You made the call. That's the most honest version of "evidence over perception" I can build: the product makes it hard to lie to yourself, and it never grades you.

The new thing you get to see

There's a gift hiding in here. Once you measure both shapes, you can finally see your deep-to-wide ratio — how much of your week was immersion versus orchestration.

In the AI era that might be the single most useful thing to know about your own work. A week that's gone 90% wide might mean you've become a beautifully leveraged builder directing a fleet of agents. Or it might mean you've quietly stopped doing the hard, lonely, deep thinking and you're just managing the machine. The tool won't tell you which — that's yours to judge. But no other tool can even show you the ratio, because they're all still measuring the broken proxy, or not measuring anything at all.

What this protects, and what it doesn't

I want to be clear that this is not "trydeepwork now rewards multitasking." Depth is still the whole point. We're not letting distraction sneak in the back door — we're fixing a definition that had quietly let distraction in the front door (the single-tasked, half-present session) while locking out genuine engagement that happened to be wide.

And there's a tension I'm not going to pretend away: an instrument-not-judge philosophy sits awkwardly next to the parts of the product that do grade you — the levels, the celebrations, the "perfect week." For this new dimension, we're going pure instrument: no confetti, no badge, no "that counted," just the reading. Whether that lens should eventually be turned on the rest of the product is a real and bigger question. Today, the call is small and clean: wide work gets measured, never graded.

Where this leaves us

trydeepwork is going to stop asking "did you do one thing?" and start honoring the question it was always really asking: "were you locked in?" — across both shapes of focus, deep and wide. It'll capture the orchestration work that GenAI made central and that the tool used to throw away. It'll keep every subjective signal consequence-free, so the mirror never lies and never judges. And it'll show you a ratio nobody else can.

I started today annoyed that two good hours were invisible. I'm ending it convinced the proxy that made them invisible was a relic of a world that no longer exists. The work changed. The tool should tell the truth about the work — all of it — and then get out of the way and let you decide what it meant.