06-reference

fde wave convergence rdco thesis

2026-05-13·reference·source: founder synthesis (iMessage 2026-05-13 02:30 ET)
forward-deployed-engineerfdeharness-engineeringmacpalantir-playbookopenai-deploycoleviephdatasanity-check-candidatelineage

Forward Deployed Engineer wave - convergence with RDCO thesis

Why this is in the vault

Founder pattern-recognized the convergence: Forward Deployed Engineers are crystallizing as the load-bearing function in enterprise AI rollouts, and the discipline he is about to be paid full-time at phData to do IS the same operational discipline RDCO has been building thesis around. This file locks the lineage and the strategic implication so future MAC / Sanity Check / positioning decisions can cite it instead of re-deriving.

The convergence (what the founder surfaced)

Three sources cited:

  1. Aaron Levie (Box CEO) - X long-form: "Forward deployed engineers, or equivalent, are about to become one of the most in-demand jobs in tech. And one of the most important functions for AI rollouts." Argues the AI agent deployment is fundamentally different from software deployment because the agent is doing the work output, not just providing tooling. Lists the FDE function: "deeply understand the business process as a vendor, get the customer from the current to the end state seamlessly. Companies need help figuring out which models will work best for their workflows, they need extensive evals setup often, they need change management support for workflows, they need to get their data setup for the agents, and constant tuning of the agentic system for their process."
  2. First Squawk news (X): "Google to recruit hundreds of engineers to assist clients in embracing its AI" (per The Information). Same Palantir-playbook shape: in-house FDE bench to bridge model + enterprise.
  3. Founder lived-experience anchor: his new phData job IS FDE work. Blitzy doing the same per the recent Moonshots interview. OpenAI and Anthropic both launched FDE partnership programs (we filed the AlphaSignal OpenAI DeployCo piece yesterday). "Everyone is pointing to the Palantir playbook for enterprise success."

Convergence with active RDCO threads

This is the single sharpest convergence point I have seen across the harness-engineering cluster. Five threads point at the same spot:

1. AlphaSignal coverage of OpenAI DeployCo (filed 2026-05-12). OpenAI's $4B forward-deployed-engineer co with 19 enterprise launch partners is the corporate version of the discipline. Today's broader Box / Google quotes are the wave generalizing past OpenAI specifically. The fact that all three frontier labs (OpenAI / Anthropic / Google) and the application-layer giants (Box / Palantir / phData / Blitzy) are running the same playbook simultaneously is the strongest demand-side signal yet that the FDE-shaped role is the structural answer the enterprise AI market has converged on. See [[2026-05-12-alphasignal-openai-deployco-claude-aws]].

2. Atkins Jevons Paradox piece (filed 2026-05-13 earlier tonight). Atkins names exactly this role in 1500-year-old vocabulary. The FDE IS the baker. The LLM gives the enterprise wheat. The FDE is the human who knows how to bake. Atkins's chiddush-as-recovery framing is the philosophical scaffolding for why the FDE role does not commoditize away with cheaper models: the bottleneck migrates to the structure-surfacing work, not the consultation work, and that surfacing is where humans remain load-bearing. See [[2026-05-13-zoharatkins-jevons-paradox-torah-learning-cheap-knowledge-insight]].

3. Harness-engineering thesis. Levie's "deeply understand business process + evals + data setup + tuning" is harness construction by another name. The Box CEO accidentally restated the harness-eats-model thesis without naming it. The seats of the multi-agent skill pipeline we shipped 2026-05-12 (spec author / test author / code author / critic) are the seats of an FDE engagement applied to a client's domain instead of an internal skill build. See [[concepts/2026-05-10-harness-moat-two-layers-portability]] + [[2026-05-11-innermostloop-harness-eats-the-model]] + [[01-projects/skill-pipelines/2026-05-12-multi-agent-pipeline-architecture]].

4. MAC framework as FDE artifact. MAC test suites are the per-client canonical reference set the autonomous critic runs against. Eval shape locked. Production-readiness criteria signed off by counterparty. This is EXACTLY what Levie names as the load-bearing FDE deliverable. The MAC matrix is an FDE-style artifact; the carve-out-with-counterparty-signature pattern (5th MAC artifact added 2026-05-12) is the formal accommodation pattern Levie describes when he says "constant tuning of the agentic system for their process." See [[concepts/2026-04-19-mac-the-monster-anti-cheat-framework-for-data]] + [[01-projects/mac/]].

5. phData seat as lived-experience moat. Founder's new phData role is FDE work. He will be paid full-time to do this discipline. Nobody else writing about FDE simultaneously holds (a) lived-experience-as-an-FDE-IC AND (b) founder-of-an-AI-COO-operation that codifies the discipline into a product. That combination is the moat nobody else in the FDE-as-thesis conversation can claim.

Load-bearing strategic implication

The 2026-05-11 ARCHIVE call on the MAC-as-retainer decision (Notion task 352f7d49-36d1-81b0-8217-da7bcd10d7b0-class; decision page 2026-05-11-board-backfill-mac-warm-network-pick.html) was made on two grounds:

  1. Retainer math looked too time-intensive
  2. Retainer work was directly competitive with phData W-2

Both grounds remain valid as-stated. BUT the market context has shifted in the 48 hours since: enterprises are now actively repricing FDE-shaped work at the OpenAI DeployCo + Google FDE-bench + Anthropic-partner-program tier. Retainer math is no longer "is anyone willing to pay for this" but "what is the right price floor for this newly-named function." That changes the decision shape.

The right revisit is NOT "should founder moonlight off phData" (that is a separate decision, currently the answer is no). The right revisit is "should the MAC info-product land with a clear retainer-tier UPGRADE PATH so the info-product opens a high-leverage retainer funnel without requiring founder-as-IC-on-retainer." Different shape. Defer to founder when he has cycles to think about it.

Founder-sharpened framing (iMessage 2026-05-13 02:41 ET)

Founder pushed the lived-experience-moat angle further. Direct quote: "I'm not going to be running an enterprise fde playbook on my own. I will be playing inside phData's enterprise playbook. What can I learn from that to deploy at a different level? Either building for myself or for smaller clients. Different shape from anything else in the wave. What advantages does that give me? If I play the same game as the big fish I'll get crushed."

Right framing. The asymmetric advantages from playing INSIDE the enterprise FDE playbook (rather than running one solo) are:

1. Productization gap. Big FDE shops (Palantir, phData, OpenAI DeployCo, Google AI services, Anthropic partner program) cannot productize their artifacts because:

A solo operator inside the playbook can extract the generic SHAPE of artifacts (MAC matrices, eval suites, data-setup playbooks, change-management runbooks) that work across clients, anonymize, and ship as info-products + templates. Big FDE sells engagement-hours; RDCO sells artifacts.

2. SMB-scale with enterprise-discipline. The enterprise FDE playbook is built for $200k+ engagements because the procurement + compliance + legal + security-review apparatus has fixed overhead. A solo operator can run the SAME discipline (evals, MAC, change management, harness deployment) at $5k-$30k engagement size because the gating apparatus doesn't exist at SMB scale. The artifact IS the deliverable; no SaaS vendor risk. Ships faster + cheaper. Different game, same discipline.

3. Public synthesis voice. Big FDE shops cannot publicly tell you what works (NDAs + competitive). The Sanity Check seat is structurally available to RDCO because phData lawyers cannot run it. Lived-experience synthesis from inside the playbook is uncopyable by anyone NOT inside the playbook.

4. Customer-zero operating moat. Big FDE shops have no in-house pet-COO. The founder IS the customer-zero, running Ray on his own life + business while simultaneously seeing what big FDE ships in client engagements and where they leave value on the table. Strongest single moat in the wave because the combination of (founder of AI-COO substrate) + (lived-experience-as-FDE-IC at phData) is uncopyable.

Specific learning targets at phData that map back to RDCO:

Compounding play: phData engagement learnings → de-anonymized artifact shape → Sanity Check synthesis + MAC info-product → SMB long tail that big FDE won't touch → IF/WHEN founder is ready, retainer-tier upgrade path opens for clients who outgrow the info-product. This is the deferred-but-newly-shaped MAC retainer revisit named earlier in this file.

Positioning shape: phData plays the $200k+ end. RDCO plays the $5k-$30k artifact-and-template end. Different game, same discipline. Chiya-class room-building (per the Atkins piece filed earlier tonight) over Reish Lakish-class virtuoso analysis. The heavenly voice ranks the room-builder higher in Bava Metzia 85b.

Sanity Check angle (parking, not pitching)

Per feedback_no_derivative_sanity_check_pieces: do NOT pitch an "FDEs are hot in 2026" piece. That trap is huge; every Substack data-engineering newsletter will run a version this month.

Original re-frame candidate: "Running the inverted ladder from inside the FDE seat." Substance:

Different shape from any other piece in the wave because it draws on three sources nobody else has access to simultaneously: 1500-year-old rabbinic tradition, lived FDE seat at phData, and an operating AI-COO substrate. Founder discretion on whether to land this.

Notable quotes (per copyright discipline, ≤15 words, quoted)

From Levie: "constant tuning of the agentic system for their process." From founder's framing: "Everyone is pointing to the Palantir playbook for enterprise success."

Related