2026 Core Accounting Software Market Map (ERP excerpt from the Agentic Finance Stack report)
Despite the title, this issue is not the market map itself. It is the ERP/core-accounting chapter teaser for CJ Gustafson's upcoming "State of the Agentic Finance Stack 2026" report (previewed in [[2026-11-05-mostlymetrics-agentic-finance-stack-2026-preview]]), built as an argument about how agentic an ERP should be, with the report early-download CTA and a podcast plug.
Why this is in the vault
It is a vertical-operator treatment of the "where does the agent get write access" question, with named vendor positions (NetSuite: control the actions, not the door; Numeric: deterministic event-to-ledger processing with the LLM as a user of the system) that parallel our harness and permissions framing.
Mapping against Ray Data Co
The load-bearing connection is the harness thesis in [[synthesis-harness-thesis-dissent-2026-04-12]]: CJ's central worry that an agent with GL write access can rewrite company history, and NetSuite's Craig Sullivan's reply that posting is fine if controls and full visibility exist, is the same tradeoff RDCO resolves with scoped permissions and audit trails (compare the PR-only and no-autonomous-external-email rules). Numeric CEO Parker Gilbert's line that processing is 100 percent deterministic while the AI orchestrates, builds and reviews is a clean external statement of "deterministic core, agent at the edge," which matters for the credibility-for-phData-sales project ([[project_credibility_for_phdata_sales]]) because it is language a CFO buyer already uses. Weak-to-medium: it confirms a pattern more than it changes any RDCO decision. Follow-up: the full report (145 companies, 12 layers) has not shipped, so nothing here to benchmark against yet. The ERP-unbundling thread also continues [[2026-04-28-cfosecrets-finance-stack-of-the-future-unbundled-erp]].
The core argument
ERPs are under pressure from both directions. CJ argues the ledger should get more locked down as the systems around it get more autonomous, since a wiped GL is unrecoverable while a wiped subledger can be rebuilt from the GL. NetSuite's Craig Sullivan counters that GL posting itself is not the thing to restrict; humans belong at vendor payments and large approvals, and the real unsexy problem is stale role permissions, because an agent will use every bit of access it is given. CJ adds that rev rec and billing have largely left the ERP, leaving audit/SOX controls as the main remaining reason to buy a midmarket ERP, and that the agent-relevant context lives in subledgers, so being deterministic in the ERP is a choice rather than a lag.
On buying: new AI ERPs win on novelty, price and implementation service (fine for an $8M company, harder for a $1B one with an audit trail); a cheaper fallback is Sage. Building a full AI ERP means building every subledger at once, which explains the funding. Numeric reframes the ERP as a financial data platform that stores the underlying business event so it can be reprocessed (for example, re-running leases after an auditor changes a borrowing rate) and diffed. Guidewire CFO Jeff Cooper adds the job-security point: system-of-record buys are career decisions, so being wrong can outweigh being right. CJ's two endings: best-of-breed wins, or AI ERPs build their own context and subledgers. His conclusion: the moat is subledger data.
Curation section
- Early-download CTA: free pre-release copy of the report via a tally.so form; 12 layers (Core Accounting, Revenue Platform, FP&A, Close, Tax, AP, Expense and Travel, Procurement, Business Accounts and Treasury, AI Spend and ROI, AI Roles), 30+ interviews, 145 companies. Self-promo.
- Report partners (underwriters): Abacum, Airwallex, BILL, cfo.ai, NetSuite, Numeric, Pigment, Proxon, Tropic, Vic.ai. NetSuite and Numeric are also the two vendors quoted in the ERP argument.
- Run the Numbers podcast: QuantumScape CFO Kevin Hettrich on saying no through a fair prioritization process rather than judging ideas, plus capital allocation. Self-promo (CJ's own show).
- Quote: Taleb, Antifragile, on asking for what people hold in their portfolio rather than their opinions.
- Deep-fetches: none (all links were self-owned or sponsor/partner).
⚠️ Sponsorship
Primary sponsor is Tabs (billing automation), masthead "proudly powered by" plus a UTM-tagged Billing Lag Calculator CTA, continuing its standing placement from the Quote-to-Cash series ([[2026-10-01-mostlymetrics-quote-to-cash-part-1]]). Separately, the report's ten underwriting partners include NetSuite and Numeric, whose executives supply the two main expert positions in this ERP argument, so treat those quotes as vendor perspectives from paying partners, not neutral analyst input. Mostly Talent/Mostly Metrics report promos are self-promo.
Related
- [[2026-11-05-mostlymetrics-agentic-finance-stack-2026-preview]]
- [[2026-04-28-cfosecrets-finance-stack-of-the-future-unbundled-erp]]
- [[2026-03-28-cfosecrets-unbundling-the-erp-tech-legacy-iv]]
- [[2026-10-01-mostlymetrics-quote-to-cash-part-1]]
- [[synthesis-harness-thesis-dissent-2026-04-12]]
- [[project_credibility_for_phdata_sales]]