Prompts Are Dead. Skills Are the… | The VC Corner Summary

1. Key Themes

Theme 1: SKILL.md Is Becoming the Universal Standard for AI Agent Behavior

In under six months, the SKILL.md format achieved cross-vendor adoption that took other protocols years to reach — making it the foundational primitive for how AI agents encode procedural knowledge.

"A Skill written for Claude Code runs unmodified in Cursor, Copilot, OpenAI Codex, Windsurf, and Lovable. MCP got cross-vendor conformance with two years of effort. Skills got there in six months because the format is markdown plus YAML."

Theme 2: Prompts Are Being Commoditized — Proprietary Data and Distribution Are the New Moats

The era of "the prompt is the product" is over. What survives is either deep workflow integration, proprietary data feeding Skills, or owning the user surface.

"Generic Skills are markdown files anyone can copy. Skills trained on your unique data are not." "If your only moat is a clever prompt, you have a head start measured in weeks."

Theme 3: Evals and Observability Are Becoming the New Infrastructure Layer

As Skills become commodity markdown, the competitive advantage shifts downstream to measuring and proving whether those Skills actually work in production.

"'Our Skills improve task success by 34% across 12,000 production hours' is [a moat]. 'We have great Skills' is not." "Evals are the new gross margin."

Theme 4: Distribution-Layer AI Companies Are Pulling Away

Platforms that own the user surface — not the model — are accruing outsized value as Skills make their products more powerful without requiring them to own the underlying AI.

"Cursor at $29.3B. Lovable at $6.6B. Replit at $9B. Vercel at $9.3B. They own the user surface. Skills make their products more powerful without changing who owns distribution."

Theme 5: The Scope of AI Builder Platforms Is Expanding from Developers to Non-Technical Users

Lovable's Skills launch signals a deliberate bet on expanding the addressable market of AI-native builders far beyond software engineers.

"Lovable shipped Skills to 8 million non-technical founders, operators, and marketers. People who have never written a line of code and never will. That is a distribution bet nobody else in this space has made at this scale."

2. Contrarian Perspectives

Contrarian 1: Skills Beat MCP Despite Being Simpler — Simplicity Wins Standards Wars

The conventional wisdom is that robust, protocol-heavy standards win in enterprise tech. Skills disproves this: a plain markdown file outpaced a heavily engineered protocol by a factor of 4x in adoption speed.

"MCP got cross-vendor conformance with two years of effort. Skills got there in six months because the format is markdown plus YAML."

The barrier being low is precisely why adoption is high — a lesson often ignored by developers who optimize for technical sophistication over interoperability.

Contrarian 2: Prompt Management Platforms Were Always Doomed — Foundation Labs Just Made It Obvious

The market treated companies like Humanloop and PromptLayer as durable SaaS businesses. The article argues they were always structurally vulnerable to being absorbed.

"Anthropic acquired Humanloop in July 2025 and shut it down on September 8. One foundation lab absorbed an entire category by acquihiring its leader. PromptLayer and Vellum are pivoting to enterprise registries. The 'prompt is the product' thesis is done."

The evidence: entire categories collapsing via acquihire in a single transaction, not gradual competitive erosion.

Contrarian 3: Long Context Is Not a Solution — It's a Performance Problem Masquerading as One

The prevailing assumption is that larger context windows reduce the need for smart retrieval architectures. The article argues the opposite — stuffing context degrades agent performance.

"Loading everything into context at once degrades agent performance. Skills route around that by loading procedural knowledge only when the agent needs it... Forty Skills installed costs roughly 1,500 tokens at startup. Stacking is free until invoked."

3. Companies Identified

Lovable

Description: AI-native app builder for non-technical users
Why mentioned: Central case study; shipped Skills to 8M non-technical builders; extraordinary growth metrics
Quote: "Lovable crossed $400M ARR in February 2026, adding $100M in a single month with 146 employees at $2.77M revenue per head. Anton Osika told Bloomberg the company is 'pacing five months ahead of projections.'"

Anthropic

Description: Foundation AI lab, creator of Claude
Why mentioned: Originated the SKILL.md standard; set the format that became universal; aggressive M&A (Humanloop acquisition); Claude Code revenue milestone
Quote: "Anthropic published the spec as an open standard at agentskills.io in December 2025. The anthropics/skills repo crossed 117,000 GitHub stars." / "Claude Code exceeded $500M run-rate by September 2025."

Cursor

Description: AI-powered coding IDE
Why mentioned: Early adopter of Skills standard; cited as a distribution-layer winner
Quote: "Cursor at $29.3B... They own the user surface. Skills make their products more powerful without changing who owns distribution."

Vercel

Description: Frontend development and deployment platform
Why mentioned: Launched skills.sh marketplace with 34,000+ Skills in January 2026; distribution-layer winner
Quote: "Vercel ships skills.sh marketplace with 34,000+ skills."

Braintrust

Description: AI evals and observability platform
Why mentioned: Leading beneficiary of the shift toward evaluation infrastructure
Quote: "Braintrust closed $80M at $800M. When Skills are commodity markdown, the moat shifts to proving whether your Skills work in production."

Harvey

Description: Vertical AI for legal
Why mentioned: Example of a vertical AI company whose moat is data and workflow, not prompts
Quote: "Harvey ($8B)... The switching cost lives in the workflow integration and the data corpus. The prompts are irrelevant."

Glean

Description: Enterprise AI search and knowledge platform
Why mentioned: Cited as a vertical AI winner with proprietary data moat
Quote: "Glean ($7.2B at $100M ARR)... The switching cost lives in the workflow integration and the data corpus."

Decagon

Description: Vertical AI for customer support
Why mentioned: Cited as vertical AI winner protected by workflow integration and proprietary data
Quote: "Decagon ($1.5B)... The switching cost lives in the workflow integration and the data corpus. The prompts are irrelevant."

Replit

Description: Browser-based coding and AI development platform
Why mentioned: Distribution-layer winner; owns user surface
Quote: "Replit at $9B... They own the user surface."

LangChain

Description: AI orchestration and agent framework
Why mentioned: Example of a category being commoditized; forced to reposition
Quote: "LangChain explicitly repositioned as an 'agent engineering platform.' Lightweight orchestration is being absorbed into IDEs and foundation labs directly."

Microsoft

Description: Enterprise software and AI platform
Why mentioned: Adopted SKILL.md standard in Visual Studio 2026 and Copilot Cowork
Quote: "Microsoft ships Agent Skills in Visual Studio 2026 and Copilot Cowork."

OpenAI

Description: Foundation AI lab
Why mentioned: Adopted SKILL.md standard with their "Hazelnut" project
Quote: "OpenAI ships Skills, codenamed 'Hazelnut,' SKILL.md compatible."

Humanloop

Description: Prompt management platform (acquired)
Why mentioned: Cautionary tale — the defining example of the "prompt is the product" thesis collapsing
Quote: "Anthropic acquired Humanloop in July 2025 and shut it down on September 8."

Abridge

Description: Vertical AI for clinical documentation
Why mentioned: Cited alongside Harvey and Glean as a vertical AI company protected by proprietary data and workflow
Quote: "Harvey ($8B), Abridge, Decagon ($1.5B), Glean ($7.2B at $100M ARR)... The switching cost lives in the workflow integration and the data corpus."

Windsurf, Cognition, Factory, Goose, Cline

Description: Various AI coding and agent platforms
Why mentioned: Cited collectively as early adopters of the SKILL.md standard
Quote: "Windsurf, Cognition, Factory, Goose, and Cline all adopt SKILL.md."

4. People Identified

Anton Osika

Description: CEO of Lovable
Why mentioned: Quoted on Lovable's growth trajectory exceeding projections
Quote: "Anton Osika told Bloomberg the company is 'pacing five months ahead of projections.'"

Ruben Dominguez

Description: Author of The VC Corner newsletter
Why mentioned: Author of this article; provides the investment and operating framework throughout
Quote: N/A — author byline only

5. Operating Insights

Insight 1: Use the Four-Question Framework to Audit Any AI Startup's Defensibility

The article offers a crisp due diligence checklist that is equally useful for founders stress-testing their own moats:

"1. Do you have proprietary data feeding your Skills? 2. Do you own the user surface? 3. Do you have evals proving your Skills measurably improve outcomes? 4. Is your workflow integration deep enough to survive an MCP-compliant competitor? Pass on four nos. Position on three yeses. Back hard on four yeses."

Insight 2: Encode Institutional Knowledge as Skills Before a Competitor Clones Your Prompts

The article frames Skills not just as a product feature but as a mechanism for companies to encode operational processes into a durable, model-portable format before generic versions become freely available.

"Skills fill the gap between knowing what to do and knowing exactly how your team wants it done... Generic Skills are markdown files anyone can copy. Skills trained on your unique data are not."

Insight 3: Token Efficiency Is a Real Engineering Constraint — Architect Accordingly

For teams building on top of agents, the Skills architecture offers a concrete lesson in how to manage context degradation in production:

"At session startup, only the name and description load. Thirty to 100 tokens per Skill. When a request matches the description, the full body loads. Forty Skills installed costs roughly 1,500 tokens at startup. Stacking is free until invoked."

6. Overlooked Insights

Overlooked Insight 1: GitHub Stars as a Signal of Standard Adoption Velocity

The article briefly cites a very specific GitHub metric that, in context, tells a meaningful story about how quickly open-source standards can achieve legitimacy in AI:

"The anthropics/skills repo crossed 117,000 GitHub stars."

For investors and founders, tracking GitHub star velocity on open standards — not just product repos — is an underused leading indicator of which primitives are winning developer mindshare before enterprise adoption follows.

Overlooked Insight 2: PromptLayer and Vellum Are Quietly Pivoting — A Category Signal Worth Tracking

While Humanloop's shutdown gets the headline, the article notes two other prompt management companies are repositioning — a detail that deserves more attention as a signal of how entire adjacent categories are repricing:

"PromptLayer and Vellum are pivoting to enterprise registries."

This suggests a broader consolidation dynamic: the surviving players in commoditized categories are racing toward enterprise data governance use cases, which may itself become crowded quickly.