What Block knows about coordination

Block is making 2 separate bets: that a company can record enough of its work for AI to carry some coordination, and that the resulting organization needs far fewer people. The first is a mechanism a small team can test on one delayed handoff. The second is Block’s high-stakes workforce decision and should not be smuggled into the smaller test.

The transferable question is concrete: can one AI role gather the current records, prepare the decision, preserve the owner’s authority, and let the next work begin sooner? The answer must come from the accepted handoff, not Block’s confidence.

Test one handoff before copying an organization

The filed workforce decision is part of the evidence

On February 26, 2026, Block filed a Form 8-K announcing a plan to reduce its workforce by more than 40%. The company estimated $450 million to $500 million in related charges, primarily for notice periods, severance, employee benefits, and share-based awards. Block also stated that expected AI benefits might not materialize.

That was not a small process improvement. People lost jobs, and Block accepted material restructuring cost and operating risk. Any account of the coordination thesis that omits that displacement is incomplete.

Block and Sequoia chair Roelof Botha published “From Hierarchy to Intelligence” as the strategic case. The authors have an economic interest in the thesis. The memo is primary evidence of what Block intends, not independent evidence that the design will work.

The memo argues that hierarchy has historically routed information through managers because people have limited capacity to hold context across a large organization. Block proposes an internal company model built from recorded work and a customer model built from transaction and behavior signals. An “intelligence layer” would use those records to coordinate work and compose products.

The memo also preserves a human edge. It assigns people intuition, opinionated direction, cultural context, trust, ethical decisions, novel situations, and high-stakes calls. That limit is more useful to a small operator than the claim that hierarchy can disappear.

Recorded work is the precondition, not the result

Block’s mechanism depends on its remote-first record. The memo says decisions, discussions, code, designs, plans, problems, and progress already exist as artifacts. An AI system can only coordinate what it can read, distinguish, and keep current.

A small business often has the opposite condition. The decision lives in the owner’s head. The customer promise sits in email. The project status is split between a meeting, a private message, and an outdated task board. Adding AI to that environment does not create a company model. It creates another reader of conflicting records.

The founder-bottleneck guide explains the first step: separate preparation and follow-through from the accountable decision. The coordination-layer guide gives the wider mechanism. The Goldratt guide asks which delayed output is actually holding back the connected work.

One Monday brief is enough for the first test

Consider a hypothetical 15-person services firm. The owner’s Monday capacity decision depends on signed work, project status, staffing changes, overdue commitments, cash constraints, and client deadlines. The source material exists, but the owner spends Monday morning reconciling it and the delivery leads wait.

The test is not “replace management.” It is “prepare one accepted operating brief before the decision window.”

Publisher disclosure: FidelicAI publishes this analysis and sells ALEK, the role used in the bounded example. Block's filings and memo do not validate ALEK or a FidelicAI customer outcome. Judge the role through the accepted brief, correction burden, retained authority, and stop condition.

ALEK, FidelicAI’s production AI chief of staff, carries priorities, decisions, meetings, commitments, risks, and weekly reporting in one operating record. Its current connections support the calendars, meeting records, documents, decision registers, tasks, dependencies, email, and legacy files relevant to that role.

ALEK can replace human preparation and follow-through work: gathering current records, separating facts from decisions, finding missing owners, assembling the brief, carrying approved commitments into the working record, and reporting overdue work. The owner retains company direction, people decisions, priorities, external commitments, and anything that carries the owner’s authority.

Done when: the accepted brief arrives in the stated work environment, every material fact traces to an approved current source, conflicts and missing owners are visible, the owner’s decisions and approvals are recorded, and each downstream lead can identify the approved next work without another status-reconstruction meeting.

Block’s later observations are evidence about Block

Block’s Q1 2026 shareholder letter reported that production code changes per engineer were more than 2.5 times the January level by mid-April. It also reported that incidents following a production code change were down more than 70% from Q1 2025 and more than 40% from Q4 2025.

Those are company-reported observations with Block’s own definitions and economic interest. The source does not publish a causal design that isolates AI from workforce changes, process changes, selection, or other conditions. The figures describe Block engineers and Block’s internal measures. They do not estimate what will happen in a 15-person services firm.

The honest reading has 4 rows:

1. Block filed a large workforce reduction and estimated its restructuring charges.
2. Block and its investor published an organization design thesis.
3. Block later reported higher code-change activity and lower post-change incident rates.
4. The public sources do not establish that the thesis caused those observations or that another company will reproduce them.

That record is enough to make Block an important live case. It is not enough to make Block the control group for everyone else.

Coordination work and people management must stay separate

An operating brief can carry context. It cannot notice an employee withdrawing in a meeting, repair trust after conflict, coach unfamiliar judgment, negotiate a client exception, or decide whose role changes. Those are not decorative human extras. They are part of management.

Block’s own memo keeps trust, culture, ethics, novelty, and high-stakes calls with people. A small team should keep a wider human boundary until the specific handoff proves otherwise.

The multi-agent teams guide applies the same rule across AI roles: separate jobs keep separate authority and work products. Do not create an agent swarm whose coordination nobody owns. The outcome guide keeps the accepted brief separate from the later business result.

The security page supplies the source and access boundary. Grant only the records and actions the role needs. A current connection is ready for its stated role operation; exact permissions, account ownership, plan or edition requirements, review, and retention still belong in the deployment record.

Use a narrow transfer test

Copy 5 parts of Block’s question:

• recorded work as the source material;
• one coordination job with a clear downstream user;
• a current accepted state;
• humans at trust, ethics, novelty, and consequential decisions;
• production evidence that can disconfirm the bet.

Do not copy 4 conclusions:

• that hierarchy is the main problem in your company;
• that every management activity is information routing;
• that AI caused Block’s reported operating changes;
• that a workforce reduction is required for the coordination test.