Uncapped with Jack AltmanBret Taylor on AI and the Future of Software | Ep. 42
CHAPTERS
- 0:00 – 1:07
AI-driven “SaaSmageddon”: why software markets are anxious
Bret frames the "SaaS-pocalypse" as broad market anxiety about defensibility rather than an indictment of every software company. He outlines how AI changes build-vs-buy dynamics and raises questions about where enduring moats come from.
- •Public market declines reflect uncertainty about software durability, not uniform company weakness
- •AI accelerates fears that software can be built quickly, eroding traditional moats
- •Defensibility becomes the central question for investors and operators
- •This moment mirrors prior platform shifts where categories stay but winners change
- 1:07 – 3:50
Systems of record: the old moat (databases, workflows, ecosystems, switching costs)
The conversation digs into why ERP/CRM/ITSM systems historically captured most enterprise value. Bret explains how being the "anchor tenant" database creates ecosystem gravity and makes switching costly.
- •Systems of record (ERP/CRM/ITSM) = databases plus workflows
- •They anchor integrations and partner ecosystems (e.g., AppExchange), creating gravity
- •High switching costs accrue through deployments, add-ons, and ecosystems
- •Scale advantages and “no one gets fired for buying IBM” dynamics reinforce incumbents
- 3:50 – 6:32
When nobody logs in: AI agents decouple apps from the system of record
Bret argues AI agents shift the user experience away from clicking through UIs toward delegation, making systems of record feel more like back-end databases. The key strategic question becomes whether the agent layer becomes the new center of gravity.
- •Agents can execute workflows invisibly, reducing the app’s UI as the locus of value
- •Systems of record still matter, but their relative value may fall
- •New gravity may come from agents that generate outcomes (e.g., lead gen)
- •This decoupling pressure applies across CRM, ERP, and ITSM
- 6:32 – 12:34
Incumbents vs startups: platform shifts, “best of breed” windows, and the strategy tax
They explore why incumbents often lag during technology shifts despite resources. Bret describes the pendulum swing back to best-of-breed, and explains how incumbents pay a “strategy tax” from legacy products, business models, and quarterly constraints.
- •New platforms (web, mobile, now AI) favor best-of-breed entrants temporarily
- •Incumbents’ existing assets create compromises that dilute the pure new-product bet
- •Business-model transitions (license→SaaS, etc.) are operationally and financially hard
- •Quarterly scrutiny and incentive structures prevent clean pivots
- 12:34 – 15:35
Sierra’s competitive reality: market maturity, RFPs, and differentiation pressures
Jack asks whether Sierra feels like a blue-ocean opportunity or a knife fight. Bret says demand is strong but competition is intense as buyers now understand agents and run formal evaluations, pushing vendors to differentiate beyond demos.
- •Demand is not constrained; the market is large and urgent for buyers
- •Capital and competitors are abundant, often with weak differentiation
- •Shift from “what is an agent?” education to “why Sierra?” selection
- •RFP-driven buying increases head-to-head competition
- 15:35 – 17:05
Winning in regulated complexity: industrial-grade agents and fast deployments
Bret explains Sierra’s wedge: serving highly regulated, complex enterprises where reliability and compliance matter more than flashy demos. He highlights the ability to handle complex conversations and go live quickly (e.g., Cigna in two months).
- •Focus on regulated industries (banks, healthcare payers/providers) as a defensible niche
- •“Industrial-grade” agents require deep engineering beyond demo-quality prototypes
- •Speed to production is a key value proposition for large enterprises
- •Operational maturity: pairing AI expertise with domain/business execution
- 17:05 – 20:54
Outcomes-based pricing: charging for solved problems, not tokens
Bret lays out his thesis that autonomous agents create a natural path to outcomes-based pricing. He contrasts this with token-based pricing, arguing tokens are an input poorly correlated with the business value customers want.
- •Outcomes are measurable in many agent settings (issue resolved, sale made)
- •Historical analogs: impressions→CPC→pay-per-install; software may move toward outcomes
- •Token pricing misaligns incentives because tokens don’t map to value delivered
- •Applied AI should be describable without referencing models/tokens
- 20:54 – 24:22
When token pricing might persist: cost centers, coding agents, and shifting reference points
They discuss scenarios where outcome measurement is messy (e.g., software engineering) and token/usage-based pricing could remain. Bret predicts competitive benchmarks may shift from “cost of an engineer” to “cost of other agents,” but value-based differentiation will still matter.
- •Coding outcomes can be hard to define cleanly vs simple service/sales outcomes
- •In the near term, buyers compare coding agents to human engineer costs
- •Over time, comparison shifts to other agents; pricing reference points evolve
- •For revenue-driving use cases (e.g., lead gen), outcomes remain the durable metric
- 24:22 – 34:12
Support tech is evolving fast: multilingual voice, noise robustness, and commoditization cycles
Bret details the real technical edge cases still blocking perfect support agents—languages, voice quality in noisy environments, speaker detection, and interruption handling. He also describes the uncomfortable reality of building capabilities that may commoditize soon, and the shift from tech-centric to product-centric selling.
- •Edge cases: Cantonese/Tagalog coverage, noisy roadside scenarios, multi-speaker dynamics
- •Need for proprietary components (voice activity detection, speaker detection) today
- •Teams build “sandcastle” features knowing they may be obsolete in ~12–24 months
- •Market progression: technology sale cycles evolve into product/platform differentiation
- 34:12 – 37:00
From support to the full customer lifecycle: “every company needs an agent in 2027”
The discussion broadens to agents across buying, servicing, and retention, using Rocket/Redfin as an end-to-end example. Bret describes agents as the branded “front door,” coordinating with other internal agents that handle specialized back-office workflows.
- •Agents expand from support into search/browsing, transactions, servicing, retention
- •Rocket example: home search → mortgage origination → mortgage servicing via agents
- •Telecom retention use case: agents conduct large-scale negotiations
- •Vision: agents as the new website—ubiquitous, branded, and lifecycle-spanning
- 37:00 – 38:47
Why generic agent builders commoditize: value accrues to agents that do specific jobs
Bret argues horizontal agent-building tools will become commodity infrastructure (foundation models, open source frameworks). Enduring businesses will be built around vertical agents that deliver concrete outcomes (support, legal review, audits, vendor onboarding).
- •Enterprise incumbents’ first AI move—agent builders—likely becomes undifferentiated
- •Expect strong tools from OpenAI and others plus open source (LangChain/LangGraph)
- •Durable value: vertical agents sold for what they accomplish, not how they’re built
- •Examples: Sierra (customer lifecycle), Harvey (legal/antitrust), future finance/procurement agents
- 38:47 – 41:46
Codex and software engineering’s flip: new best practices and the CI/CD analogy
As OpenAI board chair, Bret says the trajectory was expected—but the impact felt visceral once he used Codex. He compares the shift to the rise of true CI/CD: teams will develop new operating norms, and the first to internalize them will move fastest.
- •Emotional moment: realizing code generation is high-quality, not “slop”
- •Recent months show step-function improvement for real engineering tasks
- •Analogy: CI/CD required compatible processes; AI-native engineering will too
- •Competitive advantage comes from discovering new team best practices early
- 41:46 – 46:38
Will AI create tiny mega-companies? Competition, reinvestment, and limits of “bits-only” automation
Bret predicts some very small, very valuable companies will exist, but competition will push AI efficiency gains into reinvestment, better products, and customer acquisition. He also cautions that much of the economy is physical, constraining how far pure intelligence gains translate into headcount collapse.
- •AI efficiencies in competitive markets become reinvestment, not pure headcount reduction
- •ATM analogy: automation shifts roles; branches and staffing can persist via new value creation
- •Software/finance are highly “digital,” so they may change first and most
- •Physical-world constraints (shipping, labs, trials) limit exponential absorption
- 46:38 – 51:49
Taste, identity, and being human in an AI world
They discuss whether domains like taste, branding, and culture are “immune” to intelligence. Bret argues human social dynamics remain local and relative; people will adapt quickly, treating AI as a tool while preserving identity beyond vocation.
- •Taste and status dynamics may not track raw intelligence; social context dominates
- •Personal identity can be tied to skills (e.g., coding) but shifts as tools improve
- •Optimism: humans remain competitive, status-seeking, and socially driven
- •Hope for better interfaces than phones and a more self-actualized relationship with tech
- 51:49 – 54:58
OpenAI and advertising: mission-aligned free access if done transparently
Bret defends tasteful, clearly labeled ads as compatible with OpenAI’s mission to distribute AI widely. He argues the key is maintaining the integrity of recommendations while enabling free access for people who can’t afford subscriptions.
- •Ad-supported access can scale distribution, akin to Google search’s model
- •Core concern: ensure ads don’t taint recommendations; keep labeling and separation clear
- •Mission framing: after safety, broad access is a moral and strategic obligation
- •Affordability matters—$20/month is a real barrier for many users
- 54:58 – 1:00:37
How to run a board: written memos, real discussion, and building a complementary advisory set
Bret shares concrete board-management practices from both founder and board-member perspectives. He emphasizes written documents over slide decks, the cognitive value of writing without AI assistance, and designing boards around complementary expertise and functional relationships with management.
- •Prefer pre-read written memos to maximize substance in board meetings
- •Writing clarifies thinking; don’t outsource the synthesis process to AI
- •Boards should be built for diverse strengths (safety, legal, sales, finance)
- •Best boards act as targeted advisors to executives, not just quarterly oversight
