Rosa Del Mar

Daily Brief

Issue 57 2026-02-26

Enterprise Ai Adoption Routes And Procurement Inertia

Issue 57 Edition 2026-02-26 8 min read
General
Sources: 1 • Confidence: Medium • Updated: 2026-03-02 19:43

Key takeaways

  • The main disagreement discussed is about AI disruption risk in core B2B software rather than in sectors like fintech or energy.
  • Self-driving timelines have repeatedly taken longer than predicted, and Waymo is around $350M in revenue with low-thousands of vehicles in a few cities.
  • When a stock is priced for perfection, small increases in perceived tail risk can produce large price corrections without implying the business is eliminated.
  • Claude Code added the ability to visualize or preview apps inside Claude, reducing the need for some third-party build-and-preview tooling.
  • Existing B2B software feels dated because AI-native software is improving quickly and incumbents cannot keep up.

Sections

Enterprise Ai Adoption Routes And Procurement Inertia

  • The main disagreement discussed is about AI disruption risk in core B2B software rather than in sectors like fintech or energy.
  • Defensible value increasingly concentrates in companies with deep integrations and hard-to-replicate partnerships that AI-native competitors cannot easily displace.
  • There are four routes for AI to reach enterprises: direct purchase from model providers, in-house enterprise builds, AI-integrated incumbents, and new AI-native application companies.
  • AI-integrated incumbents and new AI-native application companies will dominate enterprise AI adoption compared with direct model-provider selling or widespread in-house builds.
  • Anthropic is unlikely to directly take over enterprise security spend from endpoint/platform incumbents because enterprises will continue buying security through established security-layer vendors.

Macro Labor And Demand Transmission Disputes

  • Self-driving timelines have repeatedly taken longer than predicted, and Waymo is around $350M in revenue with low-thousands of vehicles in a few cities.
  • The macro conclusions of the cited 'ghost GDP' disruption thesis depend on an unrealistically rapid roughly two-year adoption cycle across many industries.
  • The 'ghost GDP' concern is that productivity gains accrue to fewer humans (or to software agents) and therefore do not translate into broad consumer spending because agents buy nothing.
  • A revenue-generating investment team shrank from 12 people to 2 people while still producing eight figures of annual revenue.
  • The main risk case for AI-driven productivity is short-term disruption if displacement happens faster than workers can transition to new jobs.

Ai Feature Announcements As Risk Repricing Events

  • When a stock is priced for perfection, small increases in perceived tail risk can produce large price corrections without implying the business is eliminated.
  • Anthropic's security review feature release triggered a public-market selloff that wiped roughly $20B of market value from major cybersecurity stocks including Cloudflare and CrowdStrike.
  • After the correction, CrowdStrike was priced at about 16x revenue while projecting roughly 22% revenue growth and about 31% cash margins.
  • AI-based code security auditing and penetration-style testing were already available via Claude integrations before the market selloff.

Deployment And Product Scope Constraints For Agents

  • Claude Code added the ability to visualize or preview apps inside Claude, reducing the need for some third-party build-and-preview tooling.
  • Public-company agents have struggled because agents remain highly custom and require extensive training, onboarding, and data cleansing plus forward-deployed technical personnel that many customers and vendors cannot staff or afford at scale.
  • Horizontal platforms with many vertical use cases face weaker agents because heterogeneous workflows make it difficult to build one agent that performs well across customers.
  • Claude will implement any functionality that can be delivered inside the browser or desktop app surface, compressing defensible territory of adjacent products over time.

Agent Layer Value Migration And Saas Degradation

  • Existing B2B software feels dated because AI-native software is improving quickly and incumbents cannot keep up.
  • As AI layers absorb workflow intelligence, many SaaS apps risk becoming commodified systems of record that fail to capture incremental value.
  • As an agentic layer captures more workflow value, incumbent SaaS vendors can enter terminal decline even if they are not fully replaced.

Watchlist

  • Jason predicts a major software leader may execute a sudden 50% headcount cut and cites a Claude-generated scenario estimating a $600–$900B GDP hit, 4–5M jobs lost with multiplier effects, and severe local impacts in tech hubs if tech headcount fell 50%.
  • High-multiple momentum names like Palantir are flagged as having blow-up risk as valuation compresses (described as moving from about 70x to about 46x revenue).
  • The move is framed as notable because Jack Altman is claimed to have returned capital to LPs to join Benchmark, implying brand/platform value can outweigh solo-GP economics in the current venture environment.

Unknowns

  • Did the Anthropic security-review release lead to measurable customer budget shifts or competitive displacement for major cybersecurity vendors, beyond a short-term market repricing?
  • What are the real-world accuracy, false-positive/false-negative rates, and compliance acceptability of AI-driven security auditing compared with traditional methods in regulated enterprises?
  • How quickly are agent deployments becoming productized (lower services burden), and what implementation time and staffing levels are required today by vendor category?
  • How is enterprise AI spend splitting among direct model-provider purchases, incumbent SaaS embedding, and AI-native application vendors over the next 12–24 months?
  • To what extent will foundation-model products expand their surfaces into adjacent workflows in a way that materially reduces demand for specialized third-party tools?

Investor overlay

Read-throughs

  • Enterprise AI spend may route mainly through incumbent SaaS and AI-native apps, with direct model-provider selling and in-house builds secondary due to procurement and deployment friction.
  • Cybersecurity valuations can reprice sharply on AI feature announcements that raise perceived tail risk, even without clear near-term revenue impact, especially for high-multiple momentum names.
  • Agent deployments may scale slower than headlines suggest because customization, data readiness, and forward-deployed staffing remain bottlenecks, limiting near-term ROI and broad workflow displacement.

What would confirm

  • Customer discussions and vendor disclosures show AI budgets booked primarily as add-ons within incumbent SaaS and via AI-native apps, while direct model contracts remain smaller and concentrated.
  • Security vendor commentary indicates measurable deal pressure after AI security announcements, including longer sales cycles, heightened bake-offs, or pricing concessions tied to AI-driven auditing claims.
  • Case studies show agent implementations requiring significant services effort, extended timelines, and heavy customer data preparation, with limited productized deployments across heterogeneous workflows.

What would kill

  • Evidence emerges of enterprises buying foundation-model products directly at scale and reducing spend on incumbent SaaS or AI-native applications, overcoming procurement inertia and integration barriers.
  • Post-announcement results show cybersecurity demand and pricing remain steady with no detectable budget shift, displacement, or win-rate change attributable to AI security feature releases.
  • Vendors demonstrate broadly deployable agents with low customization, minimal forward-deployed staffing, fast time-to-value, and repeatable rollouts across varied enterprise workflows.

Sources