Jack Dorsey on Turning Companies into Mini-AGI | AI Digest
2 Video Views·Apr 29, 2026
HIERARCHY IS OVER — Block’s founder on ditching hierarchy and building “company intelligence.”
Jack Dorsey, Co-founder of Twitter & CEO of Block on Sequoia Capital
AI Digest condenses the best long-form AI podcasts from hours to minutes.
Chapters:
0:00 Intro
0:40 Hierarchy was an info hack
1:56 Make the company legible
3:04 Block’s three-role company
4:56 The CEO becomes an architect
7:06 Not ten-x, a rebuild
8:30 The three-week gut punch
10:30 Your first board hire
11:59 Taste, learning, and losing credibility
14:37 When the roadmap stops mattering
What is even a company going forward?
Jack Dorsey says he's been feeling existential dread and optimism in the same hour,
because he thinks the basic org chart might be obsolete.
This came up on Sequoia Capital's podcast in a conversation with Dorsey and Sequoia partner
and Block board member, Roloff Botta, and it's not a philosophical rant.
Block is actively reorganizing around an intelligence layer,
and Dorsey thinks it changes what a CEO even does.
The most useful way Dorsey frames hierarchy is also the least emotional.
He's not saying managers are bad people.
He's saying hierarchy was a workaround.
From first principles, the point of hierarchy is information flow at human scale.
If you need to communicate across a lot of people,
you create layers so messages don't turn into chaos.
But Block is remote first.
Every action produces an artifact.
Slack messages, emails, pull requests, code, Google Docs, recorded meetings.
It's a trail of how the company is building, failing, and fixing.
Traditionally, you rely on people in a chain to relay that information up and down.
Dorsey's claim is that this is now the lossy part.
Not because people are lying, but because humans filter, forget, soften, or spin.
So instead of routing truth through a ladder of humans, he wants to route it through a model.
Put intelligence on top of the artifacts,
and let anyone have a conversation with the company about how the company is doing.
That's the doorway to the rest of it.
Dorsey uses a phrase that matters here.
World models.
The idea is you can model the company itself, so it behaves like an intelligence you can query.
He even calls it a mini-AGI, not because it becomes sentient,
but because a company is already an intelligence.
It's just been structured in a way that's inefficient for information flow.
Picture the quarterly board meeting.
You build a giant stack of slides and docs.
You get a small window for questions.
Now imagine every board member can query the company directly,
in real time, on their own timeline.
The meeting stops being a data transfer ritual,
and becomes time for bigger existential decisions.
The throughline is legibility.
If your company were entirely legible,
you'd stop relying on people to tell you what's happening.
And once you can query the company,
the old justification for lots of layers starts to wobble.
That wobble turns into a pretty aggressive org redesign.
Block is early in the transformation, but Dorsey gives concrete targets.
Today, he says the max depth from him to anyone else is about five layers.
He wants it down to two to three this year.
In the most ideal case, he says there's no layer at all,
and everyone reports to him.
All 6,000 people, assuming most work is going through the intelligence layer.
Then he normalizes the company down to three roles.
One is the IC, builder or operator.
Engineer, designer, salesperson, product person,
whoever is working with tools to build or run the company.
The twist is they're augmented with agents,
so one person might explore what used to require a whole team.
The durable human skill here is judgment, taste, creativity.
The second role is the DRI, the person who owns customer outcomes.
They assemble a team of ICs, set strategy,
understand what roadmap solves customer needs.
The durable skill is ownership and accountability.
The third is what we'd call managers today, but he calls them player coaches.
They build the capability of other humans by doing the work and coaching along the way.
Critically, he wants this to become an assignment, not a reporting structure.
Also, his leadership team is expected to do all three.
This isn't a vibe shift, it's a power shift,
which forces a hard question.
If you flatten the org and centralize information in a model,
what is the CEO actually for?
Dorsey describes his old CEO job in three parts,
principles and team dynamics,
decisions in context of customers and competition,
and raising the bar on execution.
In the new world, those don't disappear,
but the center of gravity changes.
The CEO's job becomes the architecture of how the company as an intelligence works
and how humans align it toward the right outcomes.
He describes a visual.
The intelligence layer is in the center.
Humans are on the edge, constantly aligning it toward customer outcomes.
Then he goes further.
He thinks a company's limiting factor is its roadmap.
Customers will soon expect to ask for a feature that isn't on the roadmap
and just have it served.
He breaks block into layers.
Capabilities.
Interfaces.
Proactive intelligence.
And world models.
Capabilities are the tools.
Issue cards.
Card acceptance.
Peer-to-peer.
Lending.
Interfaces are the products customers touch.
Square.
Cash app.
Title.
Bitkey.
Proto.
Today, interfaces reflect what the company thinks customers want.
But with proactive intelligence,
the system can prompt customers based on what it already knows.
His example is money.
Money is the most honest signal in the world.
With that, you can do something simple and valuable,
like protect cash flow.
Sequence rent.
Spotify.
Allowances.
So customers don't hit zero or go negative.
He frames it as peace of mind.
And when the customer asks for some missing inventory feature,
the system should be able to compose it in real time out of capabilities.
If it can't, that gap becomes the roadmap.
That's not AI makes decisions.
It's closer to AI makes the company queryable.
And the customer becomes the loudest signal in the room.
Dorsey keeps pushing against the framing everyone uses.
This isn't primarily a productivity story,
where everyone becomes 10 times faster.
He calls it structural.
Most of the industry treats AI as a co-pilot, an add-on.
His claim is that a company should be rebuilt with AI as the core
if your business has a deep signal that gets richer over time.
If you don't have that kind of signal,
he suggests AI becomes more of an add-on.
And if you end up looking too similar to frontier labs,
differentiation and survival get hard.
He ties this urgency to his own practice.
He says that since January,
he spent three hours every morning pushing the tools,
trying to get them to do something he didn't think was possible.
Every day, he was surprised.
Roloff-Botta adds a useful analogy.
With the right signals,
lots of small participants can lead to better outcomes than
commander control, politics, and jockeying for position.
The vision here is a system closer to ground truth,
where transparency reduces the need for layers.
It sounds almost utopian
until you hear what Block did to make room for it.
They didn't nibble.
Block laid off about 40% of employees,
and Dorsey describes it as part of the reboot.
He anchors the timing to a capability jump.
In December, he says models got noticeably better
at understanding large and legacy codebases.
People came back from holidays having played with the tools,
surprised by how capable they were.
Then leadership ran a simple but brutal thought experiment.
If you had these tools from day one,
would you build the company this way?
Everyone around the table said no.
So they rebuilt from constraints.
What is the minimal number of people needed to keep the service up 100%?
Then the minimal number to stay in compliance with regulators.
Then the minimal set needed to grow,
meet commitments to the street,
and rebuild the company as an intelligence.
They added buffer and still made mistakes.
Dorsey says that going forward,
it should be easier because more of the company will be legible,
meaning more confidence in decisions.
From aspiration to execution was under three weeks.
He didn't want to wait until they had their backs against the wall
and do serial 10 or 20% cuts every year.
He calls that demoralizing and soul-crushing.
He wanted to act ahead of it,
with more integrity for people leaving and staying.
The board process moved fast, too.
Roloff says Dorsey wrote a detailed,
principled note laying out the logic,
and that trust between board and management made speed possible.
Which tees up a quieter lesson for CEOs.
If you want to make bold, structural moves,
you start by building a board relationship that can survive stress.
Dorsey has a line that should make early founders sweat,
in a productive way.
Your first board is your investors,
and you should treat that relationship as a hire you can never fire.
In fact, he reminds you,
they can fire you.
So don't optimize for brand name logos,
optimize for the person you actually want in the room when things get weird.
He also defines what the board is really for.
The core function is to ensure the company has the right CEO.
Everything else sits under that fiduciary duty.
From Roloff's side, the advice is blunt.
Your first financing, if it comes with a board member,
should be treated as a recruiting decision,
not a financing decision.
He likes getting a very good independent board member within a year or two,
certainly by product-market fit.
Independence can have a different relationship with the founder than an investor does,
and if they have experience, it can become a mentorship.
That's the governance side of legibility, too.
When the world speeds up, you need trust and clarity,
not a last-minute group project.
The personal side of leading through this speed-up is less formal.
It's about what you can tolerate,
what you're willing to learn,
and whether you can hold a point of view
when the default is to just follow the tool output.
Dorsey's most consistent regret surprised me.
It's not a specific bad decision.
It's where I decided not to learn something.
He says he embraces mistakes,
but if he isn't actively learning from them,
that's what he regrets.
He also calls out delegating too much as a worst-case pattern,
especially at Block.
He wanted multiple CEOs,
but realized they were building something more like a holding company,
with separate cultures and execution levels.
There's a practical texture to how this new era changes daily work.
He says meetings shifted from slide decks and docks to prototypes.
People walk in with something they built,
with real or simulated data,
and can modify it in real time during the conversation.
That changes old advice about focus.
He encourages more exploration than intense focus on the last 20%,
where creativity, taste, and judgment show up.
On second acts, he has a simple principle.
Leaders need to be comfortable losing credibility with stakeholders
to do something interesting.
He tells two Block stories.
When they wanted to get into lending,
the board initially said absolutely no.
CashApp was hated internally for years,
with a tiny team defending it.
He says every day he defended it,
he lost credibility.
And he was okay with that,
because he believed he could earn it back with success.
That mindset pairs with how he thinks about mentorship.
When he became CEO of Twitter,
people pushed him toward a CEO coach.
He tried it, and felt like he learned nothing.
Then he flipped his approach.
Every person, every encounter, every problem is a mentor,
if you decide to learn from it.
Roloff offers his own CEO acronym,
Authenticity, Logic, Empathy.
Dorsey adds a trait he values in people,
the ability to reprogram their mind and assumptions constantly,
and to discern signal from noise.
He warns it'll be easy to default to what the tools suggest,
treating model output as the answer, instead of input.
He keeps coming back to one old-fashioned word
that suddenly feels like the scarce resource.
Taste.
The line that lingers from this episode isn't
AI will make you faster.
It's Dorsey saying the company's limiting factor
is its roadmap,
and that customers will soon expect software
to respond like a conversation, not a menu.
If that's true,
hierarchy stops being the main way you coordinate,
because coordination becomes a property of the system.
Your job becomes building capabilities
and an intelligence layer that's legible,
queryable, and steerable by humans with judgment.
That's also why he sounds both hopeful
and a little haunted.
If your company can't become that kind of intelligence,
what exactly is the moat,
and how long does it last?
If you want the full texture,
including how Block made a 40% cut in under three weeks
and why the board backed it,
listen to Jack Dorsey with Roloff Botta
on Sequoia Capital's podcast.
It's less about the future of AI
and more about the future of company.
As a concept.

