Building Blocks: The Future Tech Stack - CFO Secrets Mailbag
Why this is in the vault
A Mailbag-format reply to a UK manufacturing CFO ("Bob") asking whether to replace his legacy Sage 200 ERP. The Secret CFO's response goes well past the question: the future finance tech stack is unbundled, best-of-breed, and connected via APIs, not a single monolithic ERP replacement. The post is a direct extension of the Buy / Build / Borrow framework from AI-for-CFOs III, applied to legacy ERP migration. RDCO files this because (a) the unbundling argument is the buyer-side mirror of the harness-engineering thesis (composable substrate + specialist applications), (b) the response models how to read tech-stack questions through a workflow-decomposition lens rather than a vendor-comparison lens, and (c) it surfaces two new tracked-author candidates (Shaka Nestoris and "Mark from Munich") who pitched their own writing publicly in the Mailbag.
The argument
The historical bet was the unified ERP - one system of record for accounting, manufacturing, inventory, payroll, and CRM, with everyone tolerating the worst-of-each in exchange for "one source of truth." That bet is breaking. Modern API-connected best-of-breed tools beat the monolith on every functional axis. The decision frame is: decompose the workflow into atomic units, find the best vendor at each unit, and treat the integration layer as the strategic asset. "Break the problem into accounting and manufacturing... find the best option at each atomic level."
For Bob specifically: don't replace Sage 200 with another monolith. Pick a modern accounting core (Campfire-class, AI-native) and pair it with a specialist manufacturing system, integrated via API.
Operational specifics
- Bob's case: £15M UK manufacturing firm, on-prem Sage 200, evaluating replacement.
- Implicit recommendation framework: atomic-decomposition first (accounting, manufacturing, AP/AR, payroll), then vendor selection per atom.
- Mailbag side-thread: Shaka Nestoris (Africa, "financial engineer") committed to 1 hour/day, 5 days/week content creation on finance topics; Mark (Munich, 5 yr controlling -> 4 yr transformation consulting -> 3 yr country FD/CFO at EUR1xx M revenue -> 3 yr Series A CFO at EUR3M ARR) relocating to London, also writing publicly.
- Author's "30 consecutive days as proof-of-concept" content discipline benchmark.
Mapping against Ray Data Co
Against the harness-engineering thesis cluster
This is the buyer-side mirror of the harness-engineering thesis. The argument is structurally identical to [[06-reference/concepts/2026-05-10-harness-moat-two-layers-portability]]:
- Atomic decomposition before vendor selection = harness-Layer-1 universal substrate before personal-Layer-2 customization. Decompose the workflow first; then pick the best tool per node.
- Integration layer as the strategic asset = the universal harness IS the substrate, the apps are interchangeable. The Secret CFO is saying the same thing from the buyer seat: the API connectivity layer is the moat, not any individual vendor.
- Specialist > generalist at every atomic node = the inversion of "one ERP, one throat to choke." Functionally identical to the [[06-reference/2026-04-11-garry-tan-thin-harness-fat-skills]] thin-harness-fat-skills argument.
The CFO seat and the AI-builder seat are independently arriving at the same architectural conclusion. This is the second strong cross-validation of the harness-engineering thesis from CFO Secrets (the first being the AI-for-CFOs III Buy/Build/Borrow framing).
Against MAC
Indirect: the unbundling argument implicitly assumes a layer that watches data quality across all the unbundled components. That layer does not exist for most mid-market finance stacks. MAC is the data-quality substrate that makes the unbundled stack safe to operate. Worth a Sanity Check angle: "If you unbundle your ERP, you also unbundle your data-quality assumptions. Here's what fills the gap."
Against Sanity Check (voice / cadence study)
Two notes:
- Mailbag format works for high-personality writers. The Mailbag cadence answers reader questions with much shorter setup than Playbooks, but earns equal personality. SC's
/remixoutputs occupy roughly this slot but are author-driven; a true reader-Q&A cadence would unlock different reply-rate dynamics, addressed in the [[06-reference/2026-04-30-sanity-check-bet-architecture-audit]] feedback-loop gap. - Public reader-pitch surface. The Mailbag is also a discovery surface for new writers (Shaka, Mark). SC could mirror this if it ever opened a reader-question intake; the secondary effect is sourcing tracked-author candidates from the readership.
Tracked-author candidates surfaced
- Shaka Nestoris - self-described "financial engineer," Africa-based, 30-day public-writing commitment. Worth tracking; if his cadence holds, he becomes a tier-K candidate himself. Capture in candidates list.
- Mark (Munich -> London) - operator-track CFO with controllership + transformation + scale-up reps. If he starts publishing, capture URL.
Sponsorship
Summation sponsored this issue (top-of-newsletter + embedded demo CTA). Campfire is referenced in the editorial body as an example of the AI-native accounting layer Bob should consider, but is NOT the paid sponsor of this issue - separate disclosure. Worth noting: the author treats Campfire as an editorial example even when not being paid, consistent with his stated investor/user position from the AI-for-CFOs series. Sponsor entity to tag forward: Summation.
Related
- [[06-reference/2026-05-11-cfo-secrets-ai-for-cfos-series-synthesis]] - the AI-for-CFOs synthesis; this Mailbag is a direct extension of the Buy / Build / Borrow argument from Part III
- [[06-reference/concepts/2026-05-10-harness-moat-two-layers-portability]] - the harness-moat thesis this piece independently validates from the buyer seat
- [[06-reference/2026-04-11-garry-tan-thin-harness-fat-skills]] - thin-harness-fat-skills; the AI-builder analog of the unbundled-ERP argument
- [[06-reference/2026-04-30-mac-bet-architecture-audit]] - MAC bet architecture; data-quality substrate for the unbundled finance stack
- [[06-reference/2026-04-30-sanity-check-bet-architecture-audit]] - SC bet architecture; Mailbag cadence is a SC-portable lesson