"WAC*" — @Benn Stancil

Why this is in the vault

Founder shared via iMessage 2026-05-22 21:33 ET with no comment (implicit "file/read this"). Stancil coins a benchmark frame that maps directly onto the RDCO "agent-deployer thesis" and the harness-vs-skills design philosophy already load-bearing in SOUL.md and across our vault — worth filing as a named handle we can cite back to.

The core argument

Stancil argues abstract model benchmarks are becoming obsolete the same way pre-tool-era job interviews did: the right evaluation question is no longer "how good is this model" but "how much value does this product deliver versus the baseline of Claude wired into everything." He proposes WAC* (Wins Above Claude and Co.) as a riff on baseball's WAR (Wins Above Replacement) — a relative scoring frame where the replacement-level player is now a generic Claude harness with full tool access.

The implication for vertical AI products (Linear, Cursor, dbt copilots, legal tech, marketing tools) is that their moat must be measurable above this baseline, not above an isolated model. Stancil half-jokingly notes the differentiator may collapse to UI polish and CSS once the underlying capability is commoditized — i.e. the harness-and-skills layer is where value accrues, not the model itself.

Named in passing: Anthropic, OpenAI, Google, Linear, dbt, Databricks, Slack, Cursor, Cloudflare. No sponsors, no CTA.

Mapping against Ray Data Co

• Agent-deployer positioning (strong reinforce). WAC* is a generalized restatement of the agent-deployer thesis we've been tracking via Tristan Handy's "BI's Second Unbundling" and Doug Turnbull's "agents replacing the search stack." Stancil's contribution is naming the evaluation primitive — the "baseline" against which everything else gets judged is now "Claude + MCP + tools." If a vertical product can't beat that, it dies. This is the same shape as Ray's COO-agent thesis: the founder runs Claude + skills + MCP servers as the substrate, and vertical SaaS becomes a "WAC* must be positive" filter for what to buy vs. build-on-substrate.
• Harness-vs-skills frame (strong reinforce). Stancil's "the differentiator may just be CSS" is a glib version of Garry Tan's thin-harness/fat-skills argument that we already operate against. The MAC info-product and the broader RDCO content arc are precisely the "value lives in the skills you load, not the model" bet. WAC* gives us a vocabulary to pitch this to data-engineering audiences who already know baseball stats.
• Investing thesis context (medium reinforce). WAC* as a frame is bearish on standalone vertical-AI startups whose moat is "we wrapped a model" and bullish on substrate vendors (Anthropic, OpenAI) plus the deployment/harness layer above them. Consistent with the [[2026-05-21-enterprise-ai-agent-deployment-paths]] hybrid-default conclusion: substrate vendors race to own runtime, consultancies race to own deployment, the squeezed middle is "AI-wrapper SaaS."
• Sanity Check angle (worth considering). Stancil basically handed us a re-frameable acronym. A Sanity Check piece titled something like "Your dbt copilot has a WAC* problem" or "What's your product's WAC*?" would have a strong hook IF we can extend the frame with original analysis rather than restate it. Per feedback_no_derivative_sanity_check_pieces, the bar is original re-frame — see "Decision to surface" below for what that would have to be.

Related

• [[2026-04-15-dbt-ade-bench-data-agent-benchmark-stancil]] — prior Stancil reference; ADE-Bench is a concrete instantiation of the WAC*-style evaluation question (does an analytics-engineering agent beat baseline?)
• [[2026-05-03-ae-roundup-bi-second-unbundling]] — Tristan Handy's "Claude Code in 15 minutes builds a dashboard" is the empirical case study for WAC* applied to BI tools
• [[2026-05-04-dataengineeringweekly-268-agents-replacing-search]] — Doug Turnbull on agent-vs-search-stack; same pattern, retrieval domain
• [[2026-04-04-claude-code-not-replacing-data-engineers]] — Robin Moffatt's DE + AI > DE test; concrete WAC* measurement in DE work
• [[2026-04-03-gerrymandering-modern-data-stack]] — earlier Stancil framing on stack politics; WAC* is the AI-era successor argument
• [[2026-05-21-enterprise-ai-agent-deployment-paths]] — three-paths model where WAC* explains why the squeezed middle is "AI-wrapper SaaS"

Copyright

Direct quotations kept under 15 words. Paraphrased throughout per RDCO fair-use convention.