Jack Dorsey: Every Company Can Now Be a Mini-AGI
In a Nutshell
Jack Dorsey proposes restructuring companies as **mini-AGI** systems, replacing hierarchies with AI-driven intelligence layers that ingest artifacts (Slack, code, docs) for real-time querying, enabling flat orgs (2-3 levels max at Block's 6,000 employees) and customer-led roadmaps via proactive capabilities, interfaces, and world models. Block executed 40% layoffs to reboot toward three roles—IC (builders), DRI (owners), player-coach—focusing human judgment on alignment while AI handles decisions. Key CEO advice: embrace constant learning from every encounter, meditate for focus/impermanence, cultivate ALE (authenticity, logic, empathy) plus taste and rapid reprogramming amid AI's structural shift.
These notes were generated by AI and may contain inaccuracies.
Since last year, Jack Dorsey has felt existential dread and hope about what companies will become. He questions traditional structures and sees only one durable structure lasting: treating companies as mini-AGI (artificial general intelligence). This stems from fear that his company could become irrelevant soon.
Jack Dorsey, founder of Twitter and Block, is the only founder with two companies in the S&P 500. Ruf Botha, Sequoia partner and Block board member, joins. Dorsey's article "From Hierarchy to Intelligence" is a manifesto rethinking organizations from first principles: eliminate hierarchy, put AI at the center. Block is transforming its organization. Dorsey seeks feedback on the article. Second half covers CEO advice: building boards, second acts (Cash App vs. original Square), CEOing two companies, being brave vs. going with the flow.
Dorsey questions hierarchy's function: it's about information flow to many people at human scale, iterated over 2,000 years. Now, foundational moment to question it, especially communication flow. Block is remote-first; every action creates artifacts (Slack messages, emails, pull requests, code, Google docs, recorded meetings) capturing how the company works, builds, fails, makes mistakes.
Traditionally, humans in hierarchy relay this up and down. Instead, put intelligence on top: build a model from artifacts, converse with the company's intelligence. Anyone, not just CEO, gets access and understanding. Treat company as mini-AGI—it is an intelligence, but not structured efficiently or losslessly for information flow.
Technology now models the company. Everyone inputs intent (strategy) and artifacts, queries it. Opens possibilities: quarterly board meetings with docs/slides limited by time—instead, board members query company intelligence real-time, focus meeting on creative/existential decisions. Same for earnings calls/analysts with fully RAG FD info on their timeline.
Scales to any role. Company architecture determines velocity and roadmap success.
Early stages. Max depth from CEO to individual: currently 5 levels, target 2-3 this year, ideally no layers—all 6,000 report to CEO via intelligence layer, manageable.
Normalize to three roles:
- IC (individual contributor): builders/operators (salesperson, engineer, designer, product). Augmented by agents; one person does team-level work. Durable skills: judgment, taste, creativity. Largest population.
- DRI (directly responsible individual): owns customer outcomes, assembles IC teams for roadmap solving needs. Durable skills: ownership, accountability.
- Player-coach: builds capability in others by doing the work themselves (IC or DRI who coaches). Not reporting structure but assignment. Durable skills: building human capacity, coaching, empathy, soft skills. Not required to be strategic.
Rarely, one person does all three (Dorsey, leadership team expected to).
Past CEO job in three parts:
- Right principles/team dynamic (hiring/firing, values, culture, tone).
- Decisions in context of customers, trends, competition.
- Raise execution bar, push uncomfortable growth.
Future: those plus architecting company as intelligence. Humans (CEO) align it to right outcomes. Visually: company world model in center, humans on edge aligning to customer outcomes.
Company's limiting factor: its roadmap. Customers expect on-demand features. Build capabilities (tools: issue cards, card acceptance, P2P lending), interfaces (Square register/dashboard, Cash App, Tidal, Bitkey, Proto), deliver capabilities there.
Move to proactive intelligence: use signals (money most honest), prompt customers (e.g., protect cash flow, sequence payments to avoid zero/negative, build cushion/savings/wealth, peace of mind).
Customers query for missing features, compose real-time from capabilities. Gaps become roadmap, surfaced by customer use/talk.
Final layer: world model (company + customer). Key human role: judgment—align to intent, values, taste, uniqueness. CEO checkpoints human alignment circle.
Works for Block due to signals. For others (e.g., Sierra, Workday) without: still applies if business deeply understands human nature (tangible signal). Otherwise, AI as add-on. Most see AI as co-pilot augmentation, not core rebuild. If rhyming too closely with frontier labs, hard to differentiate/survive.
Since January 2025 (tools like Goose agent coding harness one month before Claude Code in Jan, out of beta May), Dorsey spent 3 hours daily pushing limits—surprised every day. Compounding incredible; shift company ahead critical. Not productivity boost, but structural shift.
Ruf Botha: Adam Smith's Wealth of Nations—right signals enable self-interested participants for optimal outcomes vs. command-control hierarchies, politics. AI system provides ground truth, eliminates layers, restores founder-era productivity (100 vs. 500 people: transparency, limited hierarchy).
Reimagine human collaboration: smaller, more productive teams. Capitalist-like: signals (customer demand) decide builds, not table-pounding.
Dharmesh: ~4 years into HubSpot, eliminated org charts/titles for 9 months (everyone disliked, reinstated). Lacked intelligent signal system then.
Theory: manager mode (pyramid, VPs decide), founder mode (flat, founder decides), Dorsy mode (circle, AI decides).
Ruf: AI communicates alignment; core team sets objective function (growth rate, gross profit/employee, NPS). Humans steer/correct. Companies have multiple founding moments—harness team ideas, no hero worship.
Jack: AI facilitates context-rich decisions. Ideally, customers make majority decisions via queries/usage dictating roadmap; judge alignment to company vision. Past: infer via research/interviews/support/Twitter. Now: conversational interface gives fidelity.
Blurs lines; since last year, dread/hope on company futures—only durable structure: mini-AGI. From irrelevance fear—what differentiates, moat, customer expectations.
OpenAI showed agency desire (contain in Mac Mini). Square sellers/Cash App users (non-tech) build bots for life management via APIs. Expectation floor risen: customers as co-builders, see company as capabilities making desires fast/easy/valuable.
It really goes back to the capability set we have, the intuition of the interface, and how intelligent our world models can be to be helpful and compose UI in real time.
Spent this morning at Rogo in New York City. For a 100-person company with existing hierarchy, should they start with their system, data, or org? How to run the Dorsey playbook?
I don't know about this Dorsey playbook thing, but we don't have it all figured out. There's an important dose of humility—we're endeavoring on this path, we believe it's right, but there's a lot to figure out.
If building a company today as an AGI intelligence, what would it look like to differentiate? Starting a company now is exciting for quick prototyping and customer delivery, but there's a valley of dread around distribution and attention amid noise. Hard to surface signal of what's truly innovative and lasting.
Distribution becomes a key differentiator. There's an event horizon where current distribution (apps, websites, traditional retail) changes. If you don't have it now, it'll be hard to fight for, but new areas will emerge—don't know what yet.
A 100-person company is probably no more than 2-3 layers deep. Now's the time to question: Do I need a hierarchy? At Square, a year in, removed titles, normalized to "lead." Talking to banks with EVPs and VPs, business card culture—ripped up business cards. Everyone a "lead of" something; longer the descriptor, farther down. Kept it—no titles, just "what do you lead?" DRi: what are you ultimately responsible for? Helped a lot.
What's fundamental to solving customer problems? Where is hierarchy getting in the way? Look at tools and information generated in work—put into intelligence, query it for 2-3x better company understanding vs. relying on people (agendas, politics, emotions). Make company entirely legible: every aspect, from data perspective not far off. Add intelligence on top, make useful and proactive. Hardest: causal determination for predictability in role models around company/customer—research bet, but imminently solvable.
Bold move: laid off 40% of employees. Ruth Porat: "If you're going to eat a shit sandwich, don't nibble." Took a big bite. What was the internal debate? How big/bold? How did Ruth pitch? Board's reaction?
December last year, models noticeably upgraded: from prototypes/greenfield to understanding large/legacy codebases like ours. Hallucination less issue, tool harnesses mature. Everyone played over holidays, surprised at capabilities. Post-holidays: Would you build the company this way with these tools? Consensus: No, not this size/structure. Been tweaking edges (GM to functional structure, cap layers at 4+4), but reboot wouldn't look like today.
Exercise: 1) Minimal people to keep service 100% up. 2) Minimal for full regulatory compliance (highly regulated, legal key). 3) Minimal to grow, fulfill Street commitments, rebuild as intelligence. Built in buffer for mistakes. Span from expiration to execution under 3 weeks.
Wanted to act ahead, not with backs against wall (public company challenges). Do with integrity/generosity for leavers/stayers, toward excellence not mediocre reaction. Daily checks: right thing? Right folks? Tight group. Board open, agreed—"yes, we should do this."
Jack wrote detailed note laying logic: principled, not reactionary, logical. Management open to conversation, not dogmatic. Evolved over 3 weeks with feedback from team/board. Enormous trust between board/management—been through a lot, shorthand for crucial decisions. Gathered multiple times, board fully supportive.
You've been on many boards. Advice for CEOs: At 100 employees, bring independents? How to build board serving company/employees/investors/customers?
Early: First board is your investors—treat as hire you can't fire (they can fire you). Pressure to pick right person you'd want at company. Young founders chase VC brand names; go for the person. Optimized for Rolof to uplevel conversation/execution, challenge.
Core board function: Ensure right CEO—that's job one, beyond committees/fiduciary duty. Build board with different perspectives, open to wild ideas (like this), rationalizable with documentation/picture of opportunity cost.
Alternative: Annual 10-20% RIFs demoralizing, soul-crushing, feels like constant losing. Don't want that—presented positively to board.
For investor board members: Treat as recruiting, not financing—bigger outcome bearing. Fan of good independent within 1-2 years, certainly by product-market fit. Different founder relationship vs. investor (who might decide founder not right). Independent as mentor, especially experienced.
Boards often built too late/rushed pre-IPO (4-person to 9, no context/history/chemistry). Tested by short-seller reports, activists, tricky financings—need dynamics/alignment/values.
Jack: Started great companies, startup founder/scale-up CEO. Brian Chesky only to found two S&P 500 companies, simultaneous CEO of two publics (anti-goal/pattern—private maybe trendier).
What did you do right/wrong? Advice for fast-growing 100-person CEO?
My only regrets are where I decided not to learn something. Embrace mistakes/bad decisions, but not actively learning—that's regret.
Worst: Delegating too much, esp. Block—set multiple CEOs structure, became holding company. C-suite for Square, Cash App—value in uniting to challenge financial network (both sides of counter). Led to differing cultures/values/execution. Consistently correct delegating too much, didn't learn fast enough.
With legible company, different equation. Future regrets: Enough entropy/intent to stay relevant? Can't imagine bigger than rebuilding as intelligence.
Meetings different now vs. year ago. Two months ago: presentations/Google docs. Now: Everyone brings prototype (simulated/real data), deeper realism than slides. Modify in real time, explore breadth. Judgment: Which thread to pull? Cost of wrong path nearing zero—tools explore fast.
Startups now parallel more, productive. Used to preach focus; now less.
Wider perspective first: Past, path exploration costly (e.g., hardware prototypes weeks). Now in an hour. Encourage exploration, focus details on chosen path. 80/20: Tools build 80%, last 20% creativity/taste/judgment. Push models beyond expected—magic/focus there. Pick to ship (roadmap), but focus on capabilities/interface/proactive intelligence/world model changes everything.
CEOs struggle with second acts (Block, Bluesky, Spiral).
Never saw as second act—had to do, interesting. Building new distracts/resourcing? Cash App was that early.
Every leader comfortable losing credibility for interesting things. Founding moments: Started card reader, pivoted to lending (sellers want capital for sales). Board said no initially—lost credibility, pushed, they said yes.
Cash App: Merchant mission "make commerce easy," built P2P like email. COO Keith Rabois (PayPal founding team) said no, "solved problem." Company hated it 2-3 years, team of 8. Mentioned <8 times in S1. Investors didn't get it. Defended daily, lost credibility—knew could earn back with success (now >half business, profitable).
Comfortable losing credibility if understand earning back, principle why important. Stake reputation.
Sequoia memo on me: Strength DGAF, weakness DGAF.
High DGAF, stick to convictions. Founders struggle.
Wavers with hate/pushback. As Twitter CEO, everyone said get CEO coach—did, learned nothing. Shifted: Every person/encounter/problem is mentor if decide so. To for it to be
Every encounter teaches something. Force yourself to write down what you learn from it every day and every encounter. Biggest regrets are deciding not to learn from it, as you'd likely repeat it. Negative feedback or credibility loss is a teaching moment—decision of whether to learn from it. This gives ownership and agency: what is this trying to tell me? What am I ignoring? What am I being stubborn about? Sometimes get to the right answer, sometimes continue in old ways and fail. This mindset turns every encounter into infinite mentors, an amazing way to approach life and challenges.
You're a learner and meditator. Should all CEOs meditate? Benefits as CEO? Mark Andreessen says he's not introspective.
Him saying that he's not introspective is very introspective.
True meditation is a physical practice, training the mind to focus on one point. Meditation retreats: 10 days, first 3 days sitting from 4:30 a.m. to 9:00 p.m. focusing on breath sensation on upper lip. Sharpens focus and trains to observe sensation without emotional or intellectual reaction. Next 7 days: scan body for sensations like pain, sitting cross-legged, can't move for 3 hours at a time, super painful. Observe pain with mindset: this isn't permanent, standing up makes it go away. Trains mind to recognize everything is impermanent, no need to suffer or attach to what's going away. Apply to whole life, every emotion, reaction, encounter. Recommend because it sharpens focus, power of observation, diminishes instinct to react immediately—see things for what they are, then choose how to act. If seen as woo-woo head in clouds, that's what you get; as physical practice to strengthen mind, that's the result.
Timeless qualities of CEOs: involved with YouTube, Instagram, Block, Unity. Acronym ALE: authenticity, logic, empathy.
- Authenticity: Not pretentious, people see who you really are, behave authentically.
- Logic: Predictable, rational, don't fly off the handle.
- Empathy: Care about the team and business, opposite of sociopath, deep empathy. These three most important to keep in mind. Dealing with humans stays the same. Amid COVID, thought world would change, but Steven Pinker said things go back largely as before. Same here: companies built differently, AI transformational, upends industries, biggest drain of moats companies have seen, but basics of dealing with people and leading remain true. New: pace of change so fast—kudos to Jack for speed on decision, easy to dither 6-12 months. Move fast.
Value someone able to reprogram mind and assumptions constantly: question own requirements, assumptions, decisions, rigidity toward past, company's past, sacred cows, competition. Go wild, then distill corpus to something manageable and articulable. Extremely valuable, more so now because things move fast. Easy to go with momentum around us, increasingly hard to break free given tools. Offloading intelligence to intelligences, default to what they suggest rather than seeing as input. Still in mode where people see AI/tools as output, not better input for better output. Daily 3-hour meditation as input to make better output. Ability to take all input, see at once, phenomenal. Discern signal from noise, cool from irrelevant. Taste real: not just knowing what looks good together, but having point of view, perspective, opinionated drive, relevance over what's out there. Critical for founders: building what didn't exist because wanted it there. Now many companies are copies of copies because easy, instead of opinionated view pushing boundaries, uncomfortable.
Thank you both for Long Strange Trip, love the article, gamechanging for companies. He's unusual homo sapien with interesting thoughts on transforming company building, rethinking hierarchy—makes sense, somewhat inevitable. Grew up at PTC in '90s: very manager mode, hierarchical, power in VPs. At HubSpot: founder mode, flat not pyramid, power in founders' hands. Dorseymode next: more like circle than flat, power in system reacting real-time to employees/customers. Getting rid of hierarchy, building brain, getting inputs right, system makes decisions hierarchy used to. On to something. Curious what you think, comment on Twitter/YouTube. Jack in early innings, looking for feedback. Appreciate you all.
Keep Sequoia Capital in your library
Save the videos and channels worth coming back to, and find them again in one place.
