Uncapped with Jack AltmanAgents in the Enterprise | Aaron Levie, CEO of Box | Ep. 3
CHAPTERS
- 0:01 – 3:00
Box’s AI thesis: unlocking underused enterprise content
Aaron explains why AI is a step-change for Box: enterprises sit on huge amounts of valuable but dormant content (contracts, invoices, HR records, marketing assets) that rarely gets queried after creation. Box sees AI as the mechanism to continuously extract insights and automate work across that content, and is reimagining its product and operating model as if starting fresh in an AI-first era.
- •Most enterprise files become “cold” after initial collaboration, despite containing ongoing value
- •AI enables querying and extracting insights from long-lived content repositories
- •Box’s scale (115k customers; deep Fortune 500 penetration) positions it to deploy AI broadly
- •Company is proactively redesigning strategy and business model for an agentic future to avoid incumbent inertia
- 3:00 – 5:25
What ‘Box agents’ do: content-centric workflows in every function
Aaron lays out a practical view of agents inside Box: specialized assistants that execute repeatable, content-heavy workflows for teams like legal, procurement, marketing, and finance. These agents run in the background, turning documents and assets into actions, summaries, checks, and automated steps.
- •Agents focus on content workflows: contract review, invoice/payment term analysis, marketing asset processing
- •Box AI Studio provides primitives for customers to create/customize these agents
- •Expectation of massive proliferation: “millions” of agents tailored by role and industry
- •Agents deliver productivity by continuously executing tasks on top of enterprise data
- 5:25 – 6:50
From single-app agents to cross-system orchestration
The conversation expands from agents within Box to a broader enterprise reality: work spans dozens of SaaS systems. Aaron describes the early emergence of protocols and patterns that will let agents pull from multiple systems (Box, Salesforce, ServiceNow, data vendors, web tools) to complete end-to-end workflows and reports.
- •Enterprise workflows span platforms: Salesforce, ServiceNow, Slack, Workday, and many more
- •Future agents must federate data and actions across systems to build full context
- •Early signs: agent SDKs/tool use and developing interoperability concepts
- •Box anticipates agents being invoked from other platforms (e.g., Salesforce/ServiceNow calling Box) and vice versa
- 6:50 – 10:31
Startups vs incumbents: where new companies can still win
Jack probes whether incumbents with data and integrations have an unbeatable advantage. Aaron outlines three startup opportunity zones: incumbents “asleep at the wheel,” innovator’s dilemma business-model constraints, and entirely new AI-native categories where no traditional software incumbent exists.
- •Incumbents can lose simply by failing to pivot despite having distribution/data
- •Innovator’s dilemma: incumbents resist models that cannibalize seat-based recurring revenue
- •AI unlocks net-new categories with no clear incumbent owner
- •Large opportunity remains even when incumbents are competent, because business model transitions are hard
- 10:31 – 12:58
AI as net-new work, not just labor replacement
Aaron argues many AI impacts won’t map neatly to “percent of jobs replaced.” Instead, AI enables work companies previously couldn’t afford or staff, leading to new budgets and new functions—similar to how AI coding tools represent largely additive spending today.
- •Replacement framing can be myopic; organizations have unmet demand for work they’d do with more capacity
- •AI often creates net-new spend categories (e.g., AI coding tools) rather than direct displacement
- •Startups can out-execute adjacent incumbents (e.g., new code tools vs older platforms)
- •AI expands what teams attempt, not only how cheaply they do existing tasks
- 12:58 – 15:04
Customer support and reallocation: efficiency shifts work ‘upstream’
Jack highlights domains like support where ‘more is better’ doesn’t apply. Aaron responds that even when AI reduces ticket-handling cost, savings often get reinvested into higher-value activities like proactive customer success, changing roles rather than simply eliminating them.
- •Some workflows are capped (support tickets), making automation more directly substitutive
- •Savings can fund proactive, higher-leverage customer success work
- •SaaS organizations are constrained by cost-driven staffing ratios (CSMs per customer, SDRs per rep)
- •Workforce transitions often occur via retraining and role progression over time
- 15:04 – 16:42
Pricing agents: labor-comped today, software-margined tomorrow
They discuss how agent pricing is currently benchmarked to human labor, enabling surprisingly high price points, but may compress toward infrastructure-plus-software margins due to competition. Aaron notes exceptions where proprietary “cornered resources” (unique data) sustain durable pricing power.
- •Competition likely pushes agent pricing down toward traditional software gross margins
- •Labor-based pricing persists only with strong defensibility (e.g., unique proprietary datasets)
- •Reference to ‘Seven Powers’ and the concept of cornered resources
- •Business models will be forced to evolve away from pure seat-based logic
- 16:42 – 17:42
The real upside: TAM expansion beyond seats via ‘AI headcount’
Aaron emphasizes the biggest economic change isn’t permanent labor-level pricing—it’s that AI breaks the seat-based ceiling. A 20-person company can effectively “employ” many more AI workers (lawyers, SDRs, marketers), dramatically increasing software spend potential.
- •Seat-based SaaS caps revenue at number of employees; AI removes that cap
- •Small teams can deploy many AI agents, expanding total spend and automation scope
- •Potential for a much larger enterprise software TAM, though the math is still emerging
- •Agents become scalable capacity, not just per-user tooling
- 17:42 – 19:20
AI hype and valuation: outcomes will diverge between winners and losers
Jack asks whether AI is overhyped short-term but underhyped long-term. Aaron reframes: markets will fund many attempts; some will fail and look overvalued, while true winners will make today’s prices look cheap—an unavoidable feature of high-energy technology transitions.
- •Difficulty of labeling the whole space as over/under-valued; dispersion matters
- •Hype drives more mispriced failures, but also accelerates discovery of real winners
- •You often need multiple attempts to learn what works in a new product area
- •Ultimately, execution and product-market fit determine whether valuations age well
- 19:20 – 20:27
Being early to cloud: Box’s path dependency and the move to SaaS purity
Aaron reflects on launching just before AWS, which forced Box to build deep infrastructure capabilities and delayed a full cloud-company posture. He highlights an early, pivotal decision: refusing on-prem deployments, sticking to multi-tenant SaaS even when it cost deals.
- •Launching pre-AWS created infrastructure ‘path dependency’ and slowed cloud transition
- •Early insistence on multi-tenant SaaS (no on-prem) was painful but strategically critical
- •Architecture decisions compound over decades, influencing agility during new waves like AI
- •Regrets largely center on moving slower than desired on projects that later proved valuable
- 20:27 – 22:04
Architecture as a moat: one version of Box enables instant AI rollout
Aaron explains how SaaS uniformity became a major advantage: every customer is on the same platform version, so new AI features can ship and work immediately across the base. He also notes post-acquisition rigor—integrating capabilities into the common platform rather than running fragmented architectures.
- •Single multi-tenant version enables rapid, universal feature activation
- •AI capabilities ‘plug in’ once enabled—no customer-by-customer upgrade treadmill
- •M&A integration discipline: acquired products must conform to core platform/file system
- •Long-term benefits of early architectural clarity show up most during fast tech shifts
- 22:04 – 24:56
Neutral layer strategy: avoiding lock-in to any one cloud or model
They discuss whether content management was destined to be dominated by hyperscalers. Aaron argues Box’s differentiation is neutrality and interoperability—customers don’t want data trapped in a vertical cloud stack, especially when model quality changes quickly across OpenAI, Google, Anthropic, etc.
- •Box positioned as a neutral layer across hyperscalers rather than a vertically integrated stack
- •Customer value: ability to adopt best-in-class models as they emerge without data migration
- •Market outcome wasn’t guaranteed; required strong product execution and broad integrations
- •AI makes neutrality more valuable because model leadership shifts rapidly
- 24:56 – 28:29
Staying motivated for 20 years: platform breadth + the AI ‘dopamine’ cycle
Jack asks how Aaron maintains energy over two decades. Aaron attributes it to enjoying building, having a platform with diverse, meaningful use cases (NASA to film studios to pharma), and a renewed surge of excitement as AI unlocks dramatic new capabilities without starting from zero.
- •Sustained motivation comes from enjoyment of building and problem-solving
- •Box’s broad applicability keeps the work fresh across industries and missions
- •AI has reignited excitement through rapid, surprising product demos
- •Incumbent advantage: building on an existing platform and customer base versus starting from scratch
- 28:29 – 36:45
Shifting politics and tech: critiques of Democrats, optimism on pro-innovation policy
In the closing segment, they discuss political realignment in tech and Aaron’s experience being publicly vocal. Aaron critiques Democratic governance (especially California’s affordability and bureaucracy), argues the party needs a policy reset, and expresses cautious optimism about pro-tech, pro-innovation voices in the current administration—while disagreeing with issues like tariffs.
- •Perception that Democrats shifted left on key dimensions; tech reactions vary by priority set
- •California as an example: immense advantages undermined by housing costs and bureaucracy
- •Democrats struggle to message around real policy failures; internal split between progressive and centrist factions
- •Aaron is optimistic about pro-tech/AI innovation policy signals, though opposed to tariffs and some other directions
