a16zSoftware Finally Eats Services - Aaron Levie
CHAPTERS
- 0:00 – 0:55
AI as a mainstream consumer tool (and why early adopters forgive flaws)
The group opens by reacting to how quickly AI has become a ubiquitous consumer technology that’s bleeding into prosumer and work use. They compare today’s AI moment to earlier tech waves where early adopters tolerated rough edges and shaped the culture of adoption.
- •AI adoption feels faster and more universal than prior consumer tech shifts
- •Early adopters are unusually forgiving of errors in brand-new tech
- •New tools create new cultural norms (like early internet/video)
- •Small AI-powered teams can feel “superhuman” even in messy early stages
- 0:55 – 2:46
Immigration policy debate: who benefits and what the system optimizes for
They examine recent immigration policy proposals and the intense knee-jerk reactions they trigger. The conversation centers on whether salary thresholds or other market mechanisms would reduce gaming of the system—or further advantage big tech.
- •Immigration policy changes provoke polarized reactions, including among VCs
- •Current system is viewed as “gamed,” favoring large firms and consultancies
- •Salary-based allocation is proposed as a way to ration limited supply
- •Concern: high salary floors could still favor Amazon/Google over startups
- 2:46 – 5:39
Salary bands, body shops, and the hidden labor-market target
The panel differentiates between high-end tech hiring and the mid-tier IT/admin roles most impacted by consultant “body shops.” They discuss minimum salary bands as a lever to prevent wage suppression and reduce arbitrage in local labor markets.
- •Lottery and compliance burdens disproportionately hurt startups
- •Consultancies/body shops saturate mid-tier IT roles in many regions
- •Minimum salary bands could protect certain wage ranges and reduce arbitrage
- •Fixating on a single number (e.g., $100K) misses the broader system design
- 5:39 – 8:08
Designing an immigration system: wages, jobs, merit, and reducing waste
They step back to define the objective function: maximize merit, avoid depressing wages, and reduce systemic overhead. The group highlights how much productivity is lost to bureaucracy and uncertainty, and why simplifying the system matters as much as any threshold.
- •Policy goals differ: wage protection vs job allocation vs merit maximization
- •Best-in-world talent is framed as a positive-sum economic strategy
- •Massive organizational cost is spent “working the system” today
- •A simplified system would reduce uncertainty and rebalance advantages away from giants
- 8:08 – 12:52
From labor markets to labor productivity: why some AI studies don’t match lived reality
Shifting to AI productivity, they contrast papers showing minimal/negative impact with anecdotes from companies seeing large gains. Aaron shares internal adoption signals and argues the biggest gains come from teams engineering differently, not just typing faster.
- •Internal adoption: meaningful share of code production assisted by AI tools
- •Self-reported gains vary widely (20–30% up to ~75%)
- •Biggest predictor may be willingness to delegate aggressively to AI (“YOLO” tasks)
- •Agentic workflows shift developers from writing code to reviewing code
- 12:52 – 19:25
What drives big gains vs plateaus: expertise, workflow design, and early-adopter dynamics
They argue AI boosts domain experts most, while non-experts can fail due to poor prompts and lack of judgment. They also note that early adopters’ tolerance for mistakes can inflate perceived productivity while still enabling real breakthroughs.
- •Expertise enables verification, error-catching, and effective prompting
- •Great teams already had high baseline productivity; AI compounds it
- •Early adopters normalize imperfect tools and build new operating rhythms
- •Non-determinism changes how organizations think about reliability and deployment
- 19:25 – 24:40
Measuring AI impact: dazzlement, “shadow productivity,” and quality vs velocity
The group explains why productivity is hard to measure: AI can be “dazzling” without improving output, and many gains are personal and untracked. They suggest productivity may show up as higher-quality work, better architectures, and less drudgery—not just faster shipping.
- •AI can create enthusiasm that’s mistaken for real output gains
- •Bottom-up usage (ChatGPT, Cursor) is difficult for enterprises to measure
- •Metrics may miss improvements in robustness, documentation, and maintainability
- •Work changes shape (compressed cycles), making old measurement frameworks inadequate
- 24:40 – 31:32
Human taste + AI tools: prosumers, professionals, and Jevons paradox
They discuss how AI monetization often comes from professionals even if casual usage is broader. Human taste remains central in creative and design workflows, and time spent may stay constant while output volume/iteration increases—echoing Jevons paradox.
- •Monetized AI usage disproportionately comes from professionals
- •Professionals use AI alongside traditional tools; taste and requirements still matter
- •AI increases iteration and control rather than purely reducing cost/time
- •New “worth $20/month” utility unlocks non-monetized but valuable personal creation
- 31:32 – 37:47
Young founders and AI-native company building: velocity as the new advantage
Aaron argues a mid-2010s lull limited new categories, but AI resets the landscape. Younger founders can now build differently—moving faster, prototyping instantly, and reaching scale with tiny teams—creating a major shift in how companies are started and run.
- •AI reopens category creation after a period of platform consolidation (Slack/Zoom era)
- •Startups can achieve “instant scale” via agents and automation
- •Company-building process changes more than it did from 1995→2005 (post-internet)
- •Velocity becomes the defining edge—analogous to internet-era acceleration, amplified
- 37:47 – 41:34
Platform shifts: startups vs incumbents (and why disruption rarely means incumbents vanish)
They frame AI as a genuine platform shift that advantages startups, while cautioning that incumbents often persist and still grow. The conversation emphasizes that disruption usually expands markets rather than cleanly replacing winners, and that behavior changes are where incumbents struggle most.
- •Platform shifts historically give startups an opening; incumbent advantages are overestimated
- •Incumbents can grow even if they miss the new “agenda-setting” layer
- •Innovator’s Dilemma applies mainly via business-model resistance, not extinction
- •New user/buyer behaviors are hardest for large orgs to adapt to quickly
- 41:34 – 49:01
AI opens non-software TAM: “software finally eats services”
They argue AI turns service work into software-like products, creating opportunities where no software incumbent exists. Startups can package domain intelligence into workflows, competing more with professional services and vertical operators than with traditional SaaS companies.
- •AI converts parts of professional services into productized “AI labor”
- •Competition is often services/vertical operators, not existing software vendors
- •Customers/disrupted parties can also be the primary users of AI tools
- •AI-native agencies and integrators can undercut incumbents while expanding output
- 49:01 – 55:54
Adoption and distribution: from consumer ubiquity to enterprise upgrade cycles
They compare AI adoption rates to early internet usage and argue AI’s distribution is unprecedented because it rides on existing smartphones and consumer habits. This sets up an inevitable pull into enterprise workflows as employees demand the same interface and leverage at work.
- •Consumer-first adoption creates pressure for enterprise integration
- •Distribution is “pre-solved” via billions of phones and existing apps
- •New workforce entrants expect AI-native workflows (school → work transition)
- •Leaders can emerge early via brand effects, though first movers may not endure
- 55:54 – 59:34
Who wins over the next decade: incumbents grow, new giants emerge, and laggards may rebound
In closing, they predict a familiar pattern: incumbents get bigger, many new large companies appear, and some incumbents fail the transition. They highlight the possibility that laggards (e.g., certain enterprise giants or infrastructure players) can use AI-driven shifts to regain relevance.
- •Outcome likely mirrors cloud/SaaS/mobile: expanding markets plus new categories
- •Some incumbents won’t transition; others may rebound dramatically
- •Agenda-setting and mindshare shift faster than revenue shifts
- •Infrastructure and supply chain players may benefit from AI “factory” buildouts
