The Three-Phase Pathway for Transforming an Incumbent

Jack Dorsey published an essay in April 2026 with Sequoia's Roelof Botha titled "From Hierarchy to Intelligence." The argument is clean: corporate hierarchy has always solved one problem — routing information through organizations too large for any one person to oversee — and AI now does that routing better than human management layers. Two months earlier, Block had cut roughly 4,000 of its 10,000 employees. The stock rose 22%. Dorsey's framing of why is the part worth keeping: "I'd rather get there honestly and on our own terms than be forced into it reactively."

"On our own terms" is the entire case for a sequenced pathway. Chosen transformation retains institutional memory, preserves customer relationships, and allows course correction. Forced transformation — the kind that happens during a financial crisis, when capital is scarce and stakeholders are panicking — does none of those. But "on our own terms" is not the same as "in one move." It unfolds in three phases.

The tension: the appeal — and the trap — of the bold stroke

The decisive-leader instinct says transformation should be a single architectural choice, announced from the top, executed fast. Block's 40% cut is exhibit A for that view, and it's not wrong — for Block. The company operates in payments software, where iteration speed is a competitive weapon and a sharp headcount reduction unlocks agility the market rewards immediately.

Now hold Block next to JPMorgan. Same underlying architecture, opposite pace. JPMorgan ran 500+ AI use cases into production with no manifesto and no dramatic cut, net headcount roughly flat at 318,000 — but a different 318,000, concentrated toward the work the institution wanted humans doing. Why the difference? Because in regulated deposit banking, a compliance failure can trigger a decade-long Federal Reserve asset cap. Autonomous failure is categorically more expensive. The lesson isn't "go fast" or "go slow." It's that the same pathway accommodates both paces — and the work is sequencing it correctly, not choosing a speed in the abstract.

The reframe: three phases, three failure modes

Transformation at scale isn't one decision. It's three phases, each with a job to do and a specific way it dies.

Phase 1 — The First Venture (months 0–18). The job: prove AI-Born viability inside your specific organization. Target a segment with a clear 10× opportunity. Assemble 5 to 8 people with a separate location, distinct branding, and a genuine three-year charter — leadership committing not to kill it at the first adverse quarter. Somewhere between months 6 and 9, something goes wrong: wrong pricing to a major customer, a misclassified ticket that reaches the press, a report with hallucinated data a client finds first. That moment is the real diagnostic. A transparent post-mortem and a redesigned system signal a learning culture. Blame and tightened approvals signal theater. The target by month 18: 10,000+ customers or $5–10M revenue, iteration half-life under two weeks.

Phase 2 — Platform Abstraction (months 18–36). The job: extract what the first venture built into reusable services. The second venture should reach customers in 6 to 9 months, not the 12 to 18 the first one took. That productivity asymmetry is the proof of platform value — and the trigger for organizational resistance. The people who built the legacy infrastructure are now watching ventures leapfrog it. This is the moment the program either finds genuine executive air cover or quietly begins to die. Leadership must invest in transition infrastructure precisely here, or the rest of the journey becomes political warfare.

Phase 3 — Proliferation (months 36–60+). The job: scale to 5–10 ventures while migrating legacy functions onto the platform. Launch 2 to 3 ventures a year. By year five: 10+ ventures generating $200–500M with 100–150 employees, and 80% of new development running on the platform. The legacy organization that started at 40,000 now runs at 5,000–10,000 — smaller, more capable, structurally different.

Figure: Three phases, three purposes — prove viability, extract the platform, then scale. Each has a failure mode that ends the journey before the next phase begins.

The mechanism: each phase produces the trigger that endangers the next

The phases aren't just sequential; they're causally linked in a way that creates their own danger. Phase 1 produces a working venture — which, in Phase 2, becomes the thing legacy infrastructure owners resist. Phase 2 produces the productivity asymmetry — which is simultaneously the proof of value and the spark for political opposition. Phase 3 produces visible legacy migration — which surfaces the workforce transition that, underfunded, triggers organizational collapse even as the technology succeeds.

This is why the recurring failure isn't technical. It's the cancellation of platform investment 18 months in, as proof points are accumulating but before they convert to P&L. The ventures often hadn't failed. The organization simply never committed to funding the foundation long enough to see it work. Every phase generates its own antibody, and the discipline of the pathway is anticipating each one before it organizes.

What to do

1. Charter the first venture to survive a bad quarter. A genuine three-year commitment, in writing, with a separate location and brand. Treat the first failure as a diagnostic, not a verdict.
2. Use the Phase 1→2 transition as your air-cover checkpoint. When ventures start leapfrogging legacy infrastructure, name the resistance and fund transition support before it becomes political warfare.
3. Measure learning velocity, not P&L, through month 36. Pre-negotiate board criteria around iteration half-life so the platform budget survives the window where proof points exist but profit doesn't.
4. Match pace to your risk surface. Block's speed fits payments; JPMorgan's deliberateness fits regulated banking. Choose the pace your failure cost dictates — both run the same three phases.

The close

IBM's arc is the one to sit with if you believe AI-driven transformation is a straight line. It shed an estimated 13,000–17,000 roles across two waves — then, in February 2026, announced it would triple entry-level hiring, having discovered that AI-native talent who can direct the systems is the actual scarce resource. The trajectory ran down, then up. That's what a multi-phase transition looks like from inside: displacement, a gap where knowledge briefly evaporates, then demand for a new kind of worker.

The pathway accommodates Block's speed and JPMorgan's caution alike. What it cannot accommodate is indecision that leaves neither the efficiency gains nor the institutional knowledge intact. Three phases. Three failure modes. The work is moving through each on purpose.