What Block Actually Changed So AI Could Move Real Work
Block’s workforce reduction announcement triggered the usual online spiral: AI did it vs AI was the excuse.
That debate doesn’t help anyone who has to make work move on Monday.
The only useful question is narrower and more operational: what changed inside Block that made AI useful at scale? Not the headline. Not the headcount math. The internal mechanics that let AI touch real workflows without turning everything into noise.
Those mechanics weren’t hidden behind the announcement. They were spelled out months earlier.
In an October 26, 2025 interview on Lenny’s Podcast, Block’s CTO at the time, Dhanji R. Prasanna, described what their “AI-native” push actually looked like in practice. It wasn’t “we rolled out a chatbot.” It was a combined shift across org structure, shared standards, and system-connected automation.
1) Centralize The AI Push
First, he framed the priority clearly: the goal was to automate Block, meaning push AI-driven automation through the company’s work, not treat AI like a side experiment. The point was doing it centrally, as a company-wide effort, instead of letting each team build its own disconnected approach.
2) Eliminate The GM Structure
Second, he pointed to a “boring” organizational change that mattered more than any single tool: Block moved from a GM-style setup (a general manager model where business units run like semi-independent companies) to a functional structure.
In the GM model, units had their own engineering and design practices. In the functional model, engineering and design were unified under single leadership. That change reduced fragmentation and made it possible to drive consistent AI adoption across teams instead of reinventing the rules in every corner of the org.
3) Standardize How Work Moves
Third, he emphasized what that reorg enabled: shared policies, shared tools, and shared definitions. He gave concrete examples like role leveling meaning the same thing across the company, and people being able to move across teams into areas of need.
Those details sound bureaucratic until you’ve tried to scale AI across an organization that can’t agree on what “done” means. Standardization is what keeps AI from turning into a hundred incompatible local hacks.
4) Build An Internal Assistant With Real System Access
Fourth, he described Goose, their internal assistant, in a way that makes the “chatbot” framing look unserious. Goose isn’t just a conversation interface. It’s a general-purpose agent that can execute multi-step work and orchestrate across tools.
The key capability wasn’t model magic. It was system access.
He explains this through MCP (Model Context Protocol), which is essentially a connector standard: a way to wrap enterprise tools so an assistant can safely query and use them. With connectors in place, the assistant can pull information from real systems, run queries, generate outputs, and push results into documents or workflows.
Put simply: the assistant can reach the evidence, not just generate text about it.
5) Enable Non-Technical Teams To Build
Fifth, the most important surprise in his account was where the biggest lift showed up: not engineering.
He described non-technical teams using these tools to build small internal systems themselves, compressing weeks of wait time into hours by eliminating long waits for internal apps teams or engineering queues. That’s a structural shift: when Risk, Legal, or Ops can build what they need safely, engineering stops acting as the bottleneck for every one-off workflow request.
6) Measure Adoption And Lead By Example
Sixth, he described adoption as a leadership behavior, not a rollout memo. Executives used the tools daily.
He also described measurement beyond anecdotes: self-reported time savings, backed by operational signals tied to throughput and output (PRs, feature throughput, and other indicators), instead of treating vibes as proof.
Taken together, his description isn’t a tool story. It’s an operating model story: centralize the push, unify the org, standardize how work moves, connect the assistant to real systems, let non-technical teams build, and measure the impact like you’d measure any other operational change.
Why Most Companies Still Can’t Replicate This
Most teams hear Block’s list and walk away thinking the hard part is the tool.
The hard part is getting an approval workflow to behave like a system instead of a collection of opinions. AI only helps when the decision rules are explicit, the evidence is reachable, and the “done” state is enforceable.
That’s the part people skip. Not because it’s mysterious, but because it’s unglamorous: writing down how decisions get made, defining what proof counts, tightening handoffs, and setting a single place where the record lives.
When those basics are missing, an assistant doesn’t create speed. It creates faster drafts that still get rejected, still get reworked, and still get lost in Slack.
The simplest way to think about it: the assistant should do the manual legwork so reviewers can spend their time deciding, not hunting. That only works if every review starts with the same ingredients: a written standard, a complete evidence set, a named approver, and a stored record.
That’s what the playbook installs. It walks through how to:
Turn tribal knowledge into a decision policy
Connect the systems where proof lives (read-only first)
Automate the first draft into a complete review packet
Enable safe self-serve building for non-technical teams
Lock a traceability gate so decisions can be reproduced on demand
It also includes a small set of control metrics so you can tell whether cycle time is actually improving, or if you just made bad work move faster.
Two Ways to Build This:
Transitioning to an intelligence-native model is a high-stakes move, and how you choose to cross the gap from "strategy" to "live workflow" depends entirely on your current bandwidth and urgency. Whether you want to lead the charge internally or need a partner to handle the heavy lifting of the initial build, here is how we can move forward:
The DIY Route: If you have the internal bandwidth to lead this transformation yourself, click the button below to download the full step-by-step blueprint.
The Execution Partnership: If you want this installed correctly without the "next sprint" delays, let’s talk. Email Hello@LifeAlignedSystems.com today or book time directly on my calendar. I’ll step in to map your specific workflows, build the templates, wire your data sources, and run the pilot with your team. I don’t just hand over a document; I set the metrics loop so you can prove the throughput gains to your leadership and keep them.
Download the blueprint if you want the system. Reach out if you want it installed. Either way, start with the workflow, not the tool.