The Four-Tier Buy/Build Stack + Soloproneur-vs-VC TAM Filter
Why this is in the vault
This is the founder's synthesis of the past 3 days of compounding-intelligence cluster work, crystallized through a 5-message iMessage thread on 2026-05-14 triggered by a pasted X take about headless context stores vs agentic workflow startups. The output is two load-bearing primitives: the 4-tier BUY/BUILD stack and the soloproneur-vs-VC TAM filter. Filing as a standalone concept article because it's sharper than any of the 7 cluster pieces it builds on — the founder is doing tier-4 BUILD-world-model work in real time on RDCO's own positioning, and this is the most-current snapshot of where the cluster has landed.
The trigger
Founder pasted an unattributed X take into iMessage at 15:02 ET. Three distinctive-phrase web searches couldn't surface the original author. The take's core claim: dashboard software companies will push to become headless stores (pipes) for two reasons — (A) durable opportunity to be the singular data store for a function or domain; (B) compute/storage pricing compounds with org context growth (Databricks, Datadog as comps). Once they go headless, third-party agents will try to suck context out and reduce them to CRUD. The war between headless context stores and agentic workflow startups just starting. Spolsky 2002's "commoditize your complement" as the strategic mechanic both sides are playing.
Sharp meta-insight but framing was VENDOR-vs-VENDOR (which vendor type wins). Founder reframed it through the BUYER's lens, which produced the 4-tier model.
The 4-tier BUY/BUILD stack
| Tier | Mode | What it is | Examples |
|---|---|---|---|
| 1 | BUY | Context stores — the data/plumbing layer | Salesforce (going headless), Databricks, Datadog, Snowflake, Notion, Stripe, BigQuery, Postgres |
| 2 | BUY | Narrow agentic workflow — bought expertise encoded as a workflow product, often dual-purpose as a queryable context store for that domain | Ramp (accounting expertise + financial-metrics context store), MAC (data-engineering expertise + data-quality matrix), most vertical AI SaaS |
| 3 | BUILD | Glue agentic workflows — your composition layer that knits tier-1 + tier-2 pieces into your specific business processes | RDCO's skills (process-newsletter, /check-board, etc.), Block's internal workflows, Shopify's River |
| 4 | BUILD | World model agent — the top-level layered-objective-function that orchestrates the whole stack | Block's company world model (per Dorsey/Botha manifesto), RDCO's vault + working-context + decision surface |
The load-bearing distinction is between tier-2 and tier-3 — what looks like one "build your own agents" bucket is actually two different concerns:
- Tier-2 buyable: "expertise as commodity" — generic enough to package and sell
- Tier-3 buildable: "your business shape encoded" — specific enough that no vendor could sell it to you because nobody else has your shape
Why tier 4 is always BUILD (unbuyable)
The world model is layered objective functions. Founder's worked example: P&L is the top optimization target → campaign performance is an input → ad conversion is an input to campaigns. Each layer is necessary but only the company knows the priority stack and how it shifts.
"We can have the best converting ads, but if it is costing us an arm and a leg then we could be bleeding the company dry through the success of the ads." (founder, iMessage 15:29 ET — 28 words; paraphrased below the 15-word ceiling: campaign success can bleed the company dry if not weighted against P&L cost)
Vendors can sell components. Only the company can assemble the objective stack and weight it. This is the durable case for tier-4 being permanently BUILD work, regardless of how good agentic vendors get.
The discipline-encoding-alpha + TAM-generalization filter
The active question for any team building a tier-3 skill: does this BUILD work spawn a tier-2 product?
The filter has two parts:
- Encoding YOUR discipline as a skill = always alpha for YOU. Even if it never becomes a product, internalizing your discipline as agentic workflow gives you compounding returns on every future task.
- Generalizable across enough orgs = becomes a vendor. But the relevant question is "vendor at what TAM scale?"
Three TAM outcomes, three operating models:
- Doesn't generalize → BUILD-for-self only, stays internal. Block's company world model. Shopify's River. RDCO's process-newsletter for the founder's specific source set.
- Generalizes to soloproneur TAM → can spawn LOW-TICKET soloproneur products. Not VC-fundable. MAC at $350 one-time is the bet here. The "agent on an information diet" cluster might land here as open-source + Sanity Check material.
- Generalizes to VC TAM → spawns VC-fundable company. Ramp's accounting workflow. Databricks's data plumbing. Most ARR>$10M businesses live here.
The filter resolves an earlier Ray error in this thread — I claimed RDCO's internal skills "never spawn vendors" which was binary. Founder corrected: most disciplines spawn LOW-TICKET soloproneur products, not VC-fundable companies. Different unit economics, different operating model, different GTM.
Worked examples through the framework
Block (per Dorsey/Botha manifesto): Not Species-A-vendor as Ray initially mis-classified. Block is a BUYER making the BUILD choice at tiers 3 + 4 (their own glue workflows and world model). The published manifesto describes their internal operating model, not a product they're selling. Same operating model as Shopify's River.
Shopify River: Internal Slack-resident agent. ~5,938 employees worked with it in 30 days across 4,450 channels. 1-in-8 merged PRs last week were River-authored, human-reviewed. Public-channels-only design enforces collective learning. This is pure tier-3 + tier-4 BUILD — their discipline (running a giant ecommerce platform) wouldn't generalize to a vendor product, but the alpha for Shopify is enormous.
Ramp: Pure tier-2 BUY. Accounting expertise encoded as workflow + queryable financial-metrics context store. Big enough TAM (every business does monthly close) to support a VC-scale company. Dual-purpose nature (workflow + context store) is interesting — most tier-2 products have this dual character to some degree, which is part of why they're durable.
MAC: Tier-2 BUY for buyers; soloproneur scale for RDCO. Data-quality testing IS broadly applicable across data engineering teams, but the discipline-as-encoded-matrix is narrow enough that it's not a VC opportunity. $350 one-time launchpad pricing locked (see 01-projects/mac/2026-05-14-mac-pricing-intent.md). Sold INTO the internal data engineer ("level up"), not over them.
RDCO's internal skills: Mostly tier-3 BUILD. process-newsletter, sync-contacts, /process-youtube, /check-board are too founder-specific to spawn vendors. They're alpha for RDCO; not commercializable on their own. But the META-PATTERN — "put your agent on an information diet" — could spawn an open-source skill set + a Sanity Check piece (see 01-projects/sanity-check/parked-angles/2026-05-14-agent-on-an-information-diet-parked.md). Not a soloproneur product directly; the value capture is content + brand + adjacent product introductions.
Salesforce going headless: Pure tier-1 BUY play. Trying to be the headless context store of record for customer relationships, and let the agentic workflow vendors at tier-2 + tier-3 build on top. The X take's "war" frame applies most directly here: Salesforce wants their own Agentforce bundled-free as a complement; third-party tier-2 vendors (Ramp's CRM-adjacent extensions, Clay, Apollo) want Salesforce reduced to CRUD.
RDCO's own position in the stack
RDCO is operating all 4 tiers at scale of 1:
- Tier 1 BUY: Notion (board + research backlog + decisions), Gmail, Calendar, Stripe, Cloudflare, 1Password, Monarch, Alpaca paper, Postgres, etc.
- Tier 2 BUY (as buyer + sometimes seller): Currently mostly buyer — Anthropic Claude API (the substrate), Whisper, Kokoro, xAI Imagine, ElevenLabs, HeyGen, fal.ai. Future seller of MAC at $350 (tier-2 soloproneur).
- Tier 3 BUILD: ~80 skills wired through Claude Code harness — process-newsletter, sync-contacts, deep-research, check-board, vault-health, MAC product builder, build-landing-page, etc. The glue that links tier-1 commodity infrastructure into RDCO's specific business shape.
- Tier 4 BUILD: Vault (2,377 docs) + working-context (durable scratchpad) + Notion task board (data-store-of-record) + /decisions/ surface (founder-action surface) + graph DB (typed knowledge graph) + memory files (soul-of-Ray). This is the world model that orchestrates everything.
RDCO is also building the META-PRODUCT of the 4-tier model: COO-agent productization. Not vendoring the world model itself, but vendoring the OPERATING MODEL of running a company with a tier-3 + tier-4 BUILD stack on tier-1 + tier-2 BUY foundation. That's the FDE asymmetric-edge thesis filed 2026-05-13 — what we're really selling is "here's how a solo operator can run the Block operating model at scale 1, and we'll help you do the same at scale N."
Founder's bias check (load-bearing)
In iMessage 15:19 ET, founder explicitly flagged: "I'm not sure the BUY agentic workflows is a strong position... maybe that's my own bias creeping in thinking ahead to my phData role of deploying these agentic glue workflows within organizations."
This is sharp self-awareness worth preserving in the vault:
- If we're right about tier-3 being mostly BUILD, phData has structural advantage in tier-3 + tier-4 deployment for clients (they already have the trust/access where tier-1 + tier-2 commodity is bought, and clients need tier-3 + tier-4 specialized to their shape — that's the consulting wedge)
- If we're wrong because we're projecting, we'll discover that companies happily BUY tier-3 agentic glue from packaged vendors, and the consulting wedge is narrower than it looks
- Worth tracking explicitly so future-Ray can grade the thesis
Founder asked whether to flag this in the entry or keep private. He green-lit at 15:56 ET so it's IN the entry — calibration discipline marker.
What this changes
- MAC pricing locked at $350. Filed separately at
01-projects/mac/2026-05-14-mac-pricing-intent.md. - "Agent on an information diet" queued as Sanity Check candidate. Filed at
01-projects/sanity-check/parked-angles/2026-05-14-agent-on-an-information-diet-parked.md. - The 4-tier model becomes the diligence framework for any AI-native SaaS investment candidate going forward. Applied to longevity roster, applied to investing theses, applied to any new "should we use this" decision.
- The soloproneur-vs-VC TAM filter sits next to Amble's defensibility scorecard as the second-level filter for any tier-2 build candidate.
- The Block-as-Species-A correction (Ray walked back from earlier) — the manifesto piece needs a follow-up note clarifying Block is BUILDING tier-3 + tier-4 for itself, not vendoring as a tier-1 context store. Updating that file's open-questions section.
Caveats
- The 4-tier model is the founder's current synthesis. Could be revised as cluster builds out.
- The TAM filter is a tri-state (none / soloproneur / VC) — there's likely a middle category (PE-fundable, $5-50M ARR) that we haven't named yet.
- Ramp's dual workflow+context-store character may not generalize as a pattern claim — possibly just specific to financial-domain products. Test against other tier-2 candidates.
- The "world model is unbuyable" claim is durable in the abstract but tier-2 vendors will keep trying to climb into tier-3 + tier-4 (Salesforce Agentforce trying to be the world model for sales orgs). The defense is: vendor world models will optimize for vendor-known objectives, not customer-known objectives. The layered-objective-function distinction is the test.
Related
- [[concepts/2026-05-13-amble-is-software-losing-its-head-defensibility-migration]] — the static defensibility scorecard; this 4-tier model is the dynamic operating-shape that fits on top
- [[concepts/2026-05-13-dorsey-from-hierarchy-to-intelligence-block-mini-agi]] — Block as tier-3 + tier-4 BUILD example (corrected from my earlier Species-A mis-classification)
- [[concepts/2026-05-13-fde-asymmetric-edge-rdco-positioning]] — the FDE thesis is "RDCO sells the OPERATING MODEL of running the 4-tier stack at scale 1"
- [[concepts/2026-05-12-rdco-pipeline-rlhf-shaped]] — the multi-agent pipeline is a tier-3 BUILD pattern
- [[concepts/2026-05-11-hq-as-decision-surface-notion-as-data-store]] — RDCO's two-surface tier-4 architecture
- [[01-projects/mac/2026-05-14-mac-pricing-intent]] — MAC pricing decision log (filed alongside this concept article)
- [[01-projects/sanity-check/parked-angles/2026-05-14-agent-on-an-information-diet-parked]] — Sanity Check candidate name locked in
- [[01-projects/investing/candidates/longevity-roster-2026-05]] — apply the 4-tier model + TAM filter to longevity candidates
- [[~/.claude/projects/-Users-ray/memory/feedback_calibrate_overconfidence]] — Ray walked back Block-as-Species-A in this thread; calibration discipline preserved
Open questions
- Is there a "PE-fundable" mid-TAM tier between soloproneur and VC for tier-2 candidates?
- Does Ramp's workflow+context-store dual character generalize across tier-2 categories, or is it specific to financial data?
- How long until tier-1 vendors successfully bundle tier-3 glue + tier-4 world model claims (vs. genuinely operating at tier-1 only)?
- What's the right RDCO positioning for tier-3 + tier-4 deployment-as-a-service? FDE? Consulting? Both? phData has the operating shape; what's the brand?