a16zHow to Reorg After AI Changes Everything | Block's Owen Jennings on the a16z Show
CHAPTERS
- 0:00 – 1:46
Why AI breaks the old “more headcount = more output” assumption
Owen frames a step-change in productivity where small teams using AI tools can outperform much larger traditional orgs. He argues that for a given roadmap, companies will need fewer engineers, designers, and PMs because AI is now deeply capable inside real, complex codebases.
- •AI enables 10–100x productivity for individuals or tiny teams
- •The historic correlation between company size and output is “broken”
- •Expect fewer people needed per product roadmap (engineering/design/PM)
- •Risk that companies without a unique edge can be “vibe coded away”
- 1:46 – 4:51
What changed at Block: early agent groundwork, then a December capability jump
Block had been investing in agentic development before the recent wave, including launching an agent harness (“Goose”). Owen describes a near-binary shift in late Nov/early Dec when frontier models became strong at navigating and modifying large, existing codebases—not just greenfield projects.
- •Jack Dorsey is described as “generally right and early,” shaping early AI bets
- •Goose launched as an internal agent harness in early 2024
- •Late Nov/early Dec brought a rapid inflection in model capability
- •Existing complex codebases became tractable for AI-driven development
- 4:51 – 6:07
The 40% RIF: driven by tooling realities, not just ZIRP-era overhiring
Owen pushes back on the narrative that the layoff was mainly a correction for 2021 hiring excess. He emphasizes the cuts were heaviest in development roles because the way software is built has fundamentally changed—“we’re not writing code by hand anymore.”
- •Block’s gross profit per employee was competitive vs peers
- •The deepest cuts were on the development side, not ops teams
- •Claim: you don’t cut development heavily without a step-change in build tech
- •Statement: “We’re not writing code by hand anymore”
- 6:07 – 8:46
How the RIF was executed: principles, safety rails, and communication
Block approached the restructure as an org redesign for the AI era rather than a purely financial mandate. They prioritized reliability, regulatory trust/compliance, and durable growth while aiming to treat departing employees with transparency and respect.
- •Org design started from: “What should the org look like now?”
- •Three core principles: reliability, compliance/trust, durable growth
- •Compliance teams were largely untouched to avoid regulatory risk
- •Generous severance, no instant access cutoffs, company-wide all-hands to explain rationale
- 8:46 – 9:08
Operating post-reorg: fewer meetings, fewer layers, back to building
After the reduction, Block dramatically cut meeting load and increased direct, frequent communication. Owen describes a leaner company with larger spans of control and a renewed focus on hands-on execution rather than coordination overhead.
- •Meetings reduced ~70–80% to free time for building
- •Weekly all-hands with Jack to keep alignment high
- •Fewer layers and larger spans increase speed of information flow
- •Cultural shift from “ambiguity and shock” to execution rhythm
- 9:08 – 11:16
From sequential work to “agent swarms”: the new day-to-day workflow
The most meaningful operational change is how work happens: many agents run in parallel, generating PRs and artifacts while humans supervise, redirect, and integrate. This model applies beyond engineering to PM, growth, and other functions as people learn to manage concurrent AI output.
- •Small teams + token budgets replace large feature teams
- •Workflow shifts from linear PR flow to parallel agent-driven tasking
- •Example: running many Goose instances; context switching across agents
- •Applies to PMs and marketers too, not just engineers
- 11:16 – 13:12
What other companies need to replicate this: substrate, tools, and readiness
Owen cautions that copying the headcount move without foundational tooling is risky. Block’s approach relied on years of internal build-out, including an agent substrate and an internal “agentic operating system” to automate deterministic workflows.
- •Not every company is positioned to follow the same path
- •Block built Goose plus additional internal tooling layers over multiple years
- •Internal “G2” system lets anyone automate deterministic workflows
- •Bigger message: fewer people per roadmap, but potentially more total things built (Jevons paradox)
- 13:12 – 15:44
AI across the org: small squads, fluid staffing, and autonomous feature building
Internally, Block reshaped into small squads (1–6 people) that can move between initiatives quickly. Development is increasingly autonomous: designers and PMs ship PRs, and an internal tool (“Builder Bot”) can build and merge features to near-completion with humans finishing the last mile.
- •Org moved from classic hierarchy to small, flexible squads
- •Squads rotate across products instead of fixed long-term team assignments
- •Designers and PMs now routinely ship code changes
- •Builder Bot can autonomously build/merge PRs to ~85–100% completion
- 15:44 – 17:20
Automation outside engineering: support, ops, risk, and compliance decisioning
Beyond software development, Block is automating queue-based work and deterministic workflows across customer support and operations. Owen expects systems to outperform large human teams over time, while acknowledging that regulated environments still require humans-in-the-loop today.
- •Chat and phone support automate a majority of inquiries
- •Queue-based ops work is being automated end-to-end where possible
- •Risk/compliance decisioning increasingly model-driven
- •Humans-in-the-loop remain important for partners and regulators (for now)
- 17:20 – 18:20
The shape of Block’s business: functionalization and ecosystem integration
Owen explains the shift from separate business units (Square/Cash App/Afterpay) toward a functional structure where engineering, design, and product are centralized. This enables shared infrastructure and cross-product features that connect the ecosystem rather than optimizing each brand in isolation.
- •Moved from separate BUs with CEOs to a functional org model
- •Shared platform teams span all of Block
- •Increasing focus on features that connect Square, Cash App, and Afterpay
- •Cash App has become the majority contributor to gross profit over time
- 18:20 – 19:51
Goose as the agent platform: model-agnostic routing powering customer products
Goose is positioned as a model-agnostic harness that can swap across many models depending on the task. Block routes automations and new AI-native products through Goose, including Money Bot (Cash App) and Manager Bot (Square).
- •Goose is model-agnostic (Anthropic/OpenAI/open source; many models available)
- •Goose underpins an ‘agentic layer’ used across internal automations
- •Money Bot: proactive assistant inside Cash App, built on Goose
- •Manager Bot: analogous Square-side capability, built on Goose
- 19:51 – 23:23
From static UI to generative UI: personalized, on-the-fly app experiences
Owen predicts a rapid transition from rigid, identical interfaces to generative UI where the product experience is dynamically created. He highlights dynamic visualizations and even AI-generated mini-apps as examples, while noting the QA and safety challenges of non-deterministic outputs at scale.
- •Generative UI will make each user’s experience meaningfully different
- •Money Bot can generate charts/visualizations on the fly
- •Manager Bot could generate a custom scheduling + messaging app for a business
- •Major challenge: QA/testing non-deterministic outputs for millions of users
- 23:23 – 27:16
Defensibility in the AI era: distribution today, deep understanding tomorrow
Owen outlines near-term moats like distribution, network effects, regulatory posture, and hardware. Long-term, he believes the strongest moat will be a company’s unique, hard-to-replicate understanding—built from proprietary signal and iterated via tight loops between insight, tools, and rapid building.
- •Near/mid-term moats: network effects, distribution, licenses/regulation, hardware
- •You can clone an app, but not tens of millions of active users
- •Long-term moat: unique ‘world model’ based on rich proprietary signal
- •Future loop: codify company context + metrics, then iterate/build continuously with agents—those who can’t articulate their edge risk being ‘vibe coded away’
