Day 67 · 2026.07.26

Attention & Energy: The Scarce Resource Is Problems You Can Actually Think Through

Topic: Attention & Energy·4 principles

"A wealth of information creates a poverty of attention." — Herbert Simon

How this differs from time management: Day 38 was the calendar, Day 46 was the body and burnout. This one is the layer in between — attention. Time adds up; attention doesn't. You have 40 hours, but only three or four slots a week for problems you can genuinely think through — and an empty slot still gets taken, by the meeting that ended without a decision, the half-open call, the thought you never closed yesterday. Attention also has a second identity: it is a public strategy signal. Your team follows where you look, not your OKR doc. Four moves: cap it, absorb the switching cost, ration it by window, and manage it out loud as a signal.

PRINCIPLE 01

The Attention Budget: Three Things in Flight, Maximum Attention Is the Binding Constraint

WIP limitprioritizationmanaging up

The Principle + The Source

Treat "problems only you can think through" as fixed slots: three at a time, maximum. A fourth isn't solved by working later; it's solved by emptying a slot first — and whoever is pushing the fourth in gets to pick which one empties.

"What information consumes is rather obvious: it consumes the attention of its recipients. Hence a wealth of information creates a poverty of attention and a need to allocate that attention efficiently." — Herbert A. Simon, Designing Organizations for an Information-Rich World (1971), pp.40–41

Scene

Setting: three things already require your own judgment — the inference-serving capacity rebuild, promo packets for two senior engineers, and a schema negotiation with the data org. Monday your VP says: "Take AI cost governance too, we need a plan in Q3."

✗ The common answer

"Sure, I'll figure out the sequencing." — now four things are in flight. Two weeks later each has moved 20%, the key call on the capacity rebuild gets improvised at the last minute, and the promo packets read like paperwork. You paid four attention taxes and collected zero credit.

✓ The rewrite

"Let me state the constraint first: three things currently need my own judgment — the capacity rebuild, two promo packets, the schema negotiation. That's judgment load, not execution load; adding people doesn't think for me."

"I'll take cost governance, but something has to leave for it to come in. Three options: (1) I downgrade the schema negotiation to 'I set the principles, Alex negotiates the details' — risk is we concede one step more than we should; (2) the capacity rebuild slips a month — risk is thinner headroom before the Q4 peak; (3) on cost governance I only do the first two weeks — problem definition and the measurement standard — and Priya writes the plan with me as reviewer. I recommend (3). Which do you want?"

Checklist

Can I name the three things holding slots right now? If not, they're already filled by unowned work.
Last time someone added work, did I offer priced options — or just say "I'll try"?
How many of the three only I can think through? Why is the rest still on my desk?

Common Mistakes

Just saying "I'm swamped." Complaints don't buy resources. Offer three options with the risk priced in — that's how the sacrificed item ends up with your boss's signature on it.
Reporting judgment load as execution load. "My week is full" sounds like a calendar problem, so you'll be told to "trim some meetings." Report judgment bandwidth, not hours.
Silently downgrading instead of declaring it. Quietly doing the schema negotiation at half depth leaves you with no defense when it goes wrong. A downgrade needs one line in writing.

This Week's Practice + A Question

Do: put one line at the top of your 1:1 doc — "This week's three slots: ___ / ___ / ___." Next time work arrives, show that line first, then offer three priced options.

Ask: if two of your three slots are firefighting, is that a team capability problem — or you refusing to let go?

PRINCIPLE 02

Attention Residue: The Switching Tax Never Shows Up on the Time Sheet Attention Residue

interrupt governancedeep workteam mechanism

The Principle + The Source

The cost of an interruption isn't the five minutes. It's that for the next half hour you're running at seventy percent of yourself. Interruption is organizational behavior, so discipline won't fix it — mechanisms will: one channel, a rotating point person, a forced close-out.

"People need to stop thinking about one task in order to fully transition their attention and perform well on another. Yet, results indicate it is difficult for people to transition their attention away from an unfinished task and their subsequent task performance suffers." — Sophie Leroy, "Why Is It So Hard to Do My Work?", Organizational Behavior and Human Decision Processes 109(2), 2009

Scene

Setting: you're writing up the design trade-offs for the capacity rebuild. An engineer pings you on Slack: "Staging deploy is broken again — is it that cert issue from last time? Can you take a look?"

✗ The common answer

You switch over, read logs, send three replies, and are back in the doc twelve minutes later. But for the next half hour you're still thinking about that certificate — the trade-off section gets more tangled and you assume the problem is hard. It's residue. You rewrite the section that night.

✓ The rewrite (the reply itself)

"I'm in a deep block until 14:30. You don't need me for this class of problem: Alex is on point this week (pinned in the channel) and has full staging access; if it is the cert, the runbook is in the wiki under 'cert-rotation'. If you judge you need a decision from me before 14:30, post a blocker in #team-oncall and it's the first thing I read when I come out."

✓ The team mechanism (announce once)

(1) On-point rotation: one person a week absorbs every "quick question for BigCat"; everyone else is protected, including you on the days you're not on point.

(2) One escalation channel: real blockers go to exactly one channel; @-mentions anywhere else are async by default. Only with this rule can you actually stop watching Slack.

(3) A two-minute forced close-out: before switching, write one line — "where I got to, what's next." Leroy's experiments show that giving the unfinished task an explicit plan sharply reduces residue. That line is the plan.

Checklist

Of this week's interruptions, how many only I could answer? The rest means missing docs or no on-point.
Can the team state in one sentence what counts as a real blocker and which channel it goes to?
Does the on-point rotation include me? (If not, I've given myself a privilege — and the team notices.)

Common Mistakes + A Note for Women Leaders

Treating response speed as a virtue. Instant replies make you the most comfortable dependency on the team and the least productive person on it. Fast response is the on-point's KPI, not yours.
Setting a boundary without an alternative path. "I'm focusing, don't ping me" reads as arrogance; "don't ping me + ping him + runbook is here" is actionable.

Female Leader's Note Research repeatedly finds women get interrupted more often in meetings, and are more readily treated as the person who is "always available" — which turns them into the org's human interface: deep time shredded, and nothing block-shaped to show at review time. Two concrete moves: (1) institutionalize "always available" into a rotation — you stop being the default interface and become "who's on point this week," which is an upgrade, not a refusal; (2) use one neutral reclaiming line — "let me finish that half sentence, then I'll take yours" — which makes the interruption visible behavior instead of something you absorb alone.

This Week's Practice + A Question

Do: keep an interruption ledger for one week — three fields each time: who, could only I answer it, what was I doing. On Friday, count the share of "not only me."

Ask: if that share is over half, is the thing to fix your boundaries — or your team's docs and delegation?

PRINCIPLE 03

Ration by Window: One-Way Doors at Peak, Two-Way Doors to a Default Rule Match Decision Type to Your Best Hours

decisionsdelegationcadence

The Principle + The Source

Your judgment quality isn't constant across a day. Irreversible decisions belong in your peak window; reversible ones shouldn't touch it at all — they should be replaced by a written "don't ask me" rule. Delegation is really just moving decisions off your attention ledger.

"Some decisions are consequential and irreversible or nearly irreversible—one-way doors—and these decisions must be made methodically, carefully, slowly, with great deliberation and consultation... Most decisions aren't like that—they are changeable, reversible—they're two-way doors." — Jeff Bezos, 2015 Letter to Shareholders (Amazon)

Two Axes: Reversibility × State Window

Decision Type × Attention Window

Peak window (your sharpest 2–3 hours)

Trough (post-meetings, evening, boarding a flight)

One-way door
irreversible

Put it hereArchitecture choice, firing, promo leveling, external commitments. Give it a scheduled slot with a pre-read — not a hallway.

The expensive errorCalling an irreversible one while depleted. Say instead: "You'll have my answer tomorrow morning." One night of delay is far cheaper than one wrong call.

Two-way door
reversible

WasteSpending your best two hours picking a CI tool version is burning money. Push this down.

Hand it offWrite the default rule: below the threshold, don't ask me — just tell me once it's done.

Every other cell tolerates slippage. The top-right doesn't — it's the one cell where delay always beats deciding now.

Scene

Setting: Friday 17:40, straight out of four back-to-back meetings. An engineer catches you in the hallway: "Do we build the cache layer ourselves or take the managed third-party one? We start Monday — can you decide?"

✗ The common answer

"Take the managed one, it's faster." — three months later it has locked you to a vendor and moved data outside the compliance boundary; migrating back is two quarters. You pushed open a one-way door in twenty seconds, in your worst window.

✓ The rewrite

"First question: if this decision is wrong, how long to reverse it?" — "We'd have to migrate data, maybe a quarter." "Then it's a one-way door, and it doesn't belong to my brain after four meetings in a hallway."

"Monday 10:00, you get 45 minutes. Before Sunday, write one page: exit cost of each option, which compliance boundary the data lands in, and what code we'd touch if we switched in three months. Monday we discuss only those three."

"And here's a standing unlock for every reversible choice in this group: if the exit cost is under two engineer-weeks, you decide — one line of conclusion in the weekly."

Checklist

Which two hours are my peak window — and are they currently booked with meetings?
Which irreversible call did I make in a trough this past month? How is it holding up?
Is my "don't ask me" threshold written down (blast radius, exit cost)?

Common Mistakes

Performing speed as leadership. Deciding on the spot is a strength on two-way doors and a gamble on one-way doors. The only test is exit cost.
Delaying without a date. "Let me think about it" is a black hole; the team fills it with guesses and spins. A delay needs a timestamp and a list of inputs.

This Week's Practice + A Question

Do: post a three-line "don't ask me" rule in the team channel — exit cost under ___, you decide and tell me after; unsure whether it's reversible, first ask "how long to reverse"; one-way doors, book my peak window and bring one page.

Ask: if you keep not writing that threshold, is it fear they'll judge wrong — or fear of no longer being needed?

PRINCIPLE 04

Your Attention Is a Public Signal: Say Out Loud "I Go Deep on Two Things This Quarter" Where You Look Is the Strategy

strategy signalsustainable pacecommunicating down

The Principle + The Source

Your team doesn't read the OKR doc. They read what you keep asking about, whose reviews you attend, which topic lights you up. So the allocation has to be announced — announce nothing and the team guesses, usually as "ignored by him = abandoned."

"How you handle your own time is, in my view, the single most important aspect of being a role model and leader." — Andy Grove, High Output Management (1983)

Scene

Setting: quarterly kickoff. You've decided to go deep on two things (capacity rebuild, cost governance) and to track the other two lines (internal tooling, docs platform) by metrics only. The owner of one of those is a senior engineer who cares a lot about being seen.

✗ The common answer

You announce nothing and just quietly stop attending the internal-tooling weekly. Three weeks later that owner tells a skip-level "feels like our line doesn't matter anymore," and two months later starts looking outside. You never intended to abandon the line — you just never said so.

✓ The rewrite (the kickoff words)

"Let me start with something nobody usually says out loud: how I'm allocating my attention this quarter. I'm going deep on two — capacity rebuild and cost governance. They're the only two one-way doors this quarter."

"For internal tooling and the docs platform, I'm switching to metrics-only, no diving into details. To be explicit: that's a decision about my bandwidth, not a downgrade of those lines — headcount and perf weighting are unchanged."

"To those two owners I commit three things: a monthly 45-minute deep session on your agenda; pull me in any time you judge you need me, and that outranks my own deep block; at quarter end I present your material for you in calibration — I won't let your results get discounted because 'I wasn't involved so I don't know.'"

"Any week you feel sidelined, say so directly and I'll adjust immediately."

Checklist: The Quarterly Attention Balance Sheet

Have I written down the two things I go deep on? Are they one-way doors, or just my favorite technology?
Does my actual calendar over the last two weeks match what I announced? (When they diverge, the team believes the calendar.)
Could I run this pace for four straight quarters? If not, it isn't a sustainable pace — it's another sprint.

Common Mistakes

Announcing what you'll focus on but not what you won't. The second half is the actual information — and the only part that draws challenge. Skip it and someone else interprets it for you.
Using "I trust you" to cover walking away. Trust without mechanisms sounds like abandonment to the person who just got de-prioritized.
Mistaking sustainable for slow. Four steady quarters out-produce two sprints plus a recovery quarter. Some quarters genuinely have no sustainable option — then say plainly that you're taking on debt, and name the repayment date.

This Week's Practice + A Question

Do: send the team a five-line email — the two things you go deep on, the two lines you'll track by metrics, the three compensating commitments for the latter, and "tell me directly if you feel sidelined."

Ask: the discomfort you feel writing "what I won't go deep on" — is it fear of disappointing people, or of admitting you can't cover everything?

Going Deeper

Three attention slots works on a small team — does it still hold for a director running 60 people?

The number of slots doesn't change; what sits in them does. A front-line TLM's slots hold concrete problems (which architecture); a director's hold system design — who should decide which class of problem. The real risk at scale isn't too few slots, it's slots quietly filled by the layer below's concrete problems: you're still in three architecture reviews, so there's no bandwidth left for org design. The test: if I don't think about this, is it that nobody can — or that nobody dares? The latter is a delegation fix, not a bandwidth one.

What about peak windows on a cross-timezone team, when your peak is their midnight?

The key move: one-way doors need synchronous discussion, not a synchronous decision. Split it into an async pre-read (they write it in their peak), one short sync in an overlap where neither side is in a trough, and async confirmation (you decide in your peak). The genuinely unsolvable case is one hour of overlap that happens to be both sides' trough — then rotate the inconvenience and say so: I get up early this month, you do next month. A publicly rotated inconvenience lasts longer than pretending it suits everyone.

Won't announcing "I only go deep on two things" read upward as a lack of ambition?

It will — if you only say the second half. Reframe for your boss: not "I'm only doing two," but "I'm concentrating judgment bandwidth on these two one-way doors; the other two lines I cover with mechanisms instead of my own hours" — then show the mechanisms. What bosses fear is never that you do less; it's that nobody is minding it. And name the political cost honestly: in a culture that treats constant availability as virtue, visible focus costs you in the short run, and the only hedge is output on those two things hard enough to justify the trade retroactively.

Does funneling interruptions into one channel sacrifice weak signals that deserve to be heard?

Yes — a real trade-off. Grumbling, undercurrents, faint intent to leave usually leak out through exactly that casual Slack message. The hedge isn't reopening every channel; it's moving collection to channels whose cadence you control: 1:1s, quarterly skip-levels, plus one explicitly preserved exception — "anything personal, any time." Attention protection targets technical interruption and should never cover people: is this information about the system (it can wait) or about a person (it goes cold if it waits)?