Uncapped with Jack AltmanBret Taylor on AI and the Future of Software | Ep. 42
CHAPTERS
Why software stocks are down: anxiety about moats, not an equal indictment
Jack frames the “SaaSmageddon” narrative: AI makes software easier to build, so investors question durability. Bret argues the market move reflects uncertainty about the future of software more than a uniform negative view of every SaaS business.
- •Public market drawdown reflects broad uncertainty vs. company-by-company fundamentals
- •AI-driven “no moats” narrative is driving investor caution
- •Durability/defensibility is the core question in this moment
Systems of record before AI: databases, workflows, and switching-cost gravity
Bret explains why ERP/CRM/ITSM systems historically captured outsized value: they became the “anchor tenant” of enterprise IT. Ecosystems, integrations, and add-ons created a solar-system effect that made switching painful and reinforced incumbents.
- •Systems of record combine databases with workflows and UI-driven human processes
- •Integrations and partner ecosystems (e.g., AppExchange) create platform gravity
- •High switching costs come from the system plus the surrounding ecosystem
- •Scale advantages: sales capacity and “no one gets fired for buying IBM” dynamics
AI agents rewire the value stack: from applications people click to databases nobody logs into
The biggest shift isn’t just cheaper software creation—it’s that agents may do the work formerly done in SaaS UIs. If users don’t log into the system of record, the system’s role may shrink toward being a backend database while agent layers capture value.
- •Agents can execute workflows directly, making traditional UI interaction less central
- •Systems of record risk becoming “just databases” behind agent experiences
- •Value may migrate from stored data to the agent that generates/actions on it (e.g., leads)
- •Renewals/build-vs-buy decisions may change as custom solutions become easier
Incumbents can still win—but platform shifts temporarily favor best-of-breed upstarts
Bret compares today’s AI transition to prior shifts like the web browser and smartphones. He expects incumbents to eventually catch up, but in the transition window, best-of-breed startups can deliver a step-function improvement and gain share before incumbents adapt.
- •Incumbents often have “a right to win” via relationships and installed base
- •New platform shifts swing buying from “best of platform” to “best of breed”
- •Upstarts can be far ahead early because the skill set is genuinely different
- •Outcome is a race: can startups scale before incumbents internalize the new tech?
Why big companies move slower: the ‘strategy tax’ of assets, business models, and incentives
Jack asks why resource-rich incumbents lag. Bret describes a “strategy tax”: legacy products, migration paths, revenue models, sales comp, and quarterly scrutiny all constrain clean-sheet execution, letting small teams out-iterate large organizations.
- •Legacy strengths become constraints during platform shifts
- •Product strategy tax: trying to reuse assets and support multiple worlds at once
- •Business model tax: licensing → SaaS-style transitions reshape revenue and incentives
- •Public-company cadence makes “just pivot” unrealistic
- •Startups avoid these anchors and can pursue a pure value proposition
Sierra’s market reality: demand is huge, competition is rising, and buyers are now sophisticated
Bret describes Sierra’s growth and the changing sales motion. Early on, Sierra had to explain what agents are and address trust concerns; now large enterprises arrive with RFPs and prior evaluation, making the process more competitive and differentiation-centric.
- •Sierra growth milestones: $100M in 7 quarters; $150M in 8 quarters
- •Market shifted from education (“what is an agent?”) to urgency (“we need this yesterday”)
- •Enterprise buyers increasingly run formal RFP processes
- •Competition intensifies as more vendors look similar; likely market “culling” ahead
How Sierra wins: industrial-grade agents for regulated complexity and fast time-to-live
Sierra focuses on regulated, complex industries where real-world deployment is hard. Bret highlights “industrial-grade” reliability and the ability to go live quickly (e.g., Cigna in two months) as key differentiators.
- •Targeting regulated/complex verticals (health insurance, banks, telecom)
- •Demos are cheap; robust production agents are hard
- •Differentiation: complex conversation handling + enterprise-grade requirements
- •Operational differentiation: rapid deployment (e.g., Cigna live in ~2 months)
- •Blending AI expertise with business/change-management execution
Pricing the agent era: outcomes-based pricing vs tokens
Bret argues autonomous agents enable pricing tied to measurable outcomes (case solved, sale made) rather than inputs like token usage. He frames this as both more aligned with customer value and a marker of “applied AI” maturity.
- •Outcome-based pricing aligns incentives with what customers actually care about
- •Token-based pricing charges an input that may not correlate with business value
- •Analogies: ads evolved from impressions → CPC → pay-per-install
- •Applied AI = value proposition explainable without mentioning models/tokens
- •Trade-off: outcome pricing can put revenue at risk, but can increase willingness to pay
When tokens might make sense: engineering tools, shifting benchmarks, and cost-center vs revenue outcomes
Bret acknowledges token/usage pricing can fit where outcomes are hard to define—especially when the customer is a technical user who understands model economics. Over time, benchmarks may shift from human labor costs to competing agents, changing how value is judged.
- •Coding outcomes can be harder to quantify than support resolutions or sales wins
- •Today’s coding tools are often compared to engineer cost; later to other agents’ costs
- •Cost-center use cases may tilt toward cost-based comparisons
- •Revenue-driving agents (e.g., lead gen) maintain clearer value-based math
- •Model progress feels uneven depending on task: “good enough” vs frontier-moving domains
Support agents’ last-mile reality: multilingual voice, noise, and building tech that will commoditize
Bret details the remaining technical challenges in voice support: language coverage, noisy environments, and interruption handling. Sierra builds proprietary components (voice activity detection, multi-speaker detection) even knowing parts will become commodities as the market matures.
- •Edge cases: Cantonese/Tagalog support, non-Western voice model gaps
- •Noise challenges: horns, kids, background chatter, interruption mis-detection
- •Need for proprietary voice activity detection and multi-speaker detection
- •Market moving from tech-centric selling to product-centric selling (like web hosting → Shopify)
- •Building “temporary” advantages is necessary to be best at each stage
What becomes durable when code is cheap: prompts, product decisions, and ‘terraforming’ software
The conversation shifts to what’s defensible when much code can be generated quickly. Bret suggests durable value may move toward the systems/prompting/decision frameworks that encode countless product choices—previously locked into code—and how teams capture those decisions.
- •Traditional IP feels less permanent when code can be regenerated quickly
- •Durable asset may shift to prompts/system design that encode product decisions
- •Most product detail historically lives in code, not PRDs
- •AI disrupts software engineering first by changing how software is made
- •Teams must adapt to rapid change without over-optimizing for transient advantages
Codex and the future of engineering teams: CI/CD analogy and new best practices
As OpenAI board chair, Bret expected big leaps in coding agents, but the emotional impact hit when he used them. He predicts new team “best practices” will emerge—like CI/CD once did—and the companies that adapt first will move much faster.
- •Coding agents felt materially different over the last ~3 months
- •Board-level awareness vs first-hand usage: the ‘it’s real’ moment
- •Analogy: CI/CD transformed teams but is hard to retrofit into legacy processes
- •New operational playbooks for AI-native engineering teams will crystallize
- •Early adopters of the new workflow will outpace slower-moving organizations
Will AI create tiny billion-dollar companies? Competition, reinvestment, and the ‘bits vs atoms’ limit
Bret thinks 10-person billion-dollar companies will exist, but won’t be the norm because competition forces reinvestment of AI-driven efficiencies into gaining share. He also argues much of the economy is physical, so digital intelligence won’t compress everything equally fast.
- •Competition drives cost savings into pricing, acquisition, and product investment (ATM analogy)
- •Headcount might shift across roles rather than simply shrink overall
- •Software/finance are highly automatable because they’re mostly digital
- •Physical-world constraints (shipping, labs, trials) slow purely exponential adoption
- •“Bits” may cheapen faster than “atoms,” at least until robotics advances
Human identity, taste, and optimism: AI as tool, status dynamics, and a push toward better interfaces
Jack asks if taste/brand/storytelling are immune; Bret argues taste isn’t purely intelligence and remains local and human. He’s optimistic people will adapt—identity detaches from tasks (like coding)—and hopes AI leads to better human-computer interfaces than phone “glowing rectangles.”
- •Taste/status are social and local; not simply replaced by smarter machines
- •Personal arc: initial shock at coding agents → rapid normalization as a tool
- •Technology has long outperformed humans physically; intelligence shift is new but manageable
- •Potential cultural counter-signal: valuing being offline
- •Hope for new interfaces beyond screens as AI becomes conversational
OpenAI, ads, and distribution: funding broad access without corrupting the experience
Responding to debate sparked by competitor comments and OpenAI advertising, Bret supports tasteful, clearly labeled ads as a way to fund free access aligned with OpenAI’s mission—after safety. He emphasizes that affordability matters and good advertising can help small businesses grow.
- •Mission framing: safety first, then wide distribution of benefits
- •Ads can enable free access at scale if clearly labeled and not distorting outputs
- •Bret’s Google/AdWords perspective: ads funded massive free utility
- •Acknowledges critiques but rejects the idea ads must “taint” the product
- •Affordability matters: $20/month is significant for many users
Financing and boardcraft at Sierra: trusted partners, written memos, and boards as advisor networks
Bret describes Sierra’s investor lineup (Benchmark, Sequoia, Greenoaks) and the relationship-driven first round with Peter Fenton. He shares how to make boards effective: written documents over slides, using writing to clarify thinking (not outsourcing to AI), and recruiting directors with complementary strengths who actively advise operators.
- •Board composition reflects staged financing and choosing both firm + person
- •Benchmark/Peter Fenton relationship enabled fast, trust-based fundraising
- •Best practice: pre-read written board docs to drive substantive discussion
- •Writing is strategic synthesis; don’t outsource it to AI if it removes clarity work
- •Design boards as a system of complementary advisors for functional leaders
What’s next: regulated-industry acceleration and a contrarian regulatory ‘agents required’ prediction
Bret expects the next year to be defined by deeper adoption in highly regulated industries as AI moves beyond early adopters. His hot take: regulators may eventually demand agent-based controls because human-only processes could be viewed as riskier than monitored AI systems.
- •Shift from early adopters to mainstream deployment in regulated sectors
- •“Hard stuff” becoming the focus: higher-stakes, more complex processes
- •Prediction: regulators may ask for agents as a control mechanism
- •Reframing risk: human-only controls could be seen as the bigger risk over time
