Post-Insight Execution: How Lakehouse Insights Return to ERP/CRM and Get Executed
Current Solutions · Vendor Landscape · Metaprise’s Position · Capability Gaps
July 28, 2026 · Metaprise Internal Strategy Research
After the Insight: The Market Landscape for Data Activation and Business Execution
Post-Insight Execution: How Lakehouse Insights Return to ERP/CRM and Get Executed
Current Solutions · Vendor Landscape · Metaprise’s Position · Capability Gaps
July 28, 2026 · Metaprise Internal Strategy Research
Methodology: four parallel deep-research streams (Reverse ETL & Data Activation / Databricks native & zero-copy alliances / iPaaS & embedded integration / the process execution layer). Key facts (M&A, GA status, pricing, executive statements) were cross-verified against multiple sources; items that could not be independently confirmed are flagged in the text. Full source list in Appendix B.
0. Executive Summary
This report answers one question: once Databricks (or any lakehouse) produces an insight, how do customers get the result back into ERP/CRM and complete the business execution — who does this today, how far do they get, where are the gaps, what can Metaprise contribute, and what do we still need to build. The ten conclusions below are ordered by strength of evidence; all key facts have been cross-verified against multiple sources (see Appendix B).
- “Moving data back into ERP/CRM” is not one problem — it is four: data sync-back (D1), transactional writeback (D2), process execution (D3), and evidence & audit (D4). The market has only solved D1, while the pain customers actually pay for sits in D2–D4 (§1).
- D1 is already commoditized: Census was absorbed by Fivetran in May 2025, and Fivetran completed its merger with dbt Labs in June 2026; the surviving pure-play, Hightouch ($2.75B valuation), has pivoted wholesale into marketing-domain AI Decisioning. Zero-copy alliances (Salesforce, SAP, ServiceNow) further devalue “moving the data” itself (§3.1).
- D2 is the thinnest layer of supply in the entire market: across the whole reverse-ETL category, the only write capability into SAP S/4HANA is Hightouch’s single “Sales Order” object; supply for transactional writes to Oracle Fusion, Workday, and Dynamics F&O is zero. ERP writeback today still equals a custom integration project (§3.1, §3.3).
- Every off-the-shelf path for D3 terminates inside a single-vendor island: Data Actions trigger Salesforce Flow, SAP BDC’s return flow lands only in the analytics layer and is handed to Joule, ServiceNow’s execution lives inside the Now platform and its licensing, and Power Automate is the Microsoft island. Cross-system, one approval model, one evidence chain, one rollback story — nobody owns it (§3.2).
- D4 has zero supply market-wide: every platform offers “run logs” only. Nobody has turned “insight → approval → action → business outcome” into an immutable, exportable, audit-acceptable evidence product (§3.4, §4).
- Databricks has formally outsourced writeback to its ecosystem: the Partner Connect reverse-ETL category has exactly two members (Hightouch, Census — both marketing-oriented); Databricks Ventures chose to invest in Hightouch rather than build; and neither DAIS 2025 nor 2026 shipped any writeback product. At the same time it is growing action-layer primitives from the inside (MCP Services approval policies, Unity AI Gateway, Genie Agents) — leaving an independent execution layer a roughly 12–18 month window (§3.2).
- The execution gap has hard data: MIT measured that 95% of GenAI pilots show no P&L impact; Gartner predicts 40%+ of agentic projects will be cancelled by end-2027 and only 17% of organizations have truly deployed; Camunda’s 2026 report: 71% are “using” agents, but only 11% reach production and 80% are Q&A rather than operational actors. The demand side is nonetheless accelerating: BCG reports CEOs plan to double AI budgets in 2026. Pain and money coexist; the timing holds (§4.4).
- No single company is selling the complete bundle: insight-triggered + graded human authority + immutable evidence + compensation/rollback + outcome-based pricing + a partner-IP revenue-share marketplace. Every fragment exists; the combination does not. Two of these are zero-supply points — an evidence-chain product and outcome pricing (ServiceNow’s president publicly says outcome pricing “nobody’s been able to make work”; Pega’s per-completed-case fee is the only adjacent precedent) (§3.4).
- Metaprise’s position is therefore not “a better integration tool” but the execution layer sitting between every lakehouse and every system of record: plug in along Databricks’ own rails (table triggers, Delta Sharing, Genie API, MCP Services, Marketplace Apps), upgrade AuditChain into an Execution Receipt evidence product, and use the Mission Store to fill the partner revenue-share white space (§5).
- The three highest priorities: transactional ERP-connector depth (attack the thinnest part of the market), Databricks-native integration + a joint Blueprint reference implementation, and productizing evidence (D4) (§6).
1. “Moving Data Back into ERP/CRM” Is Actually Four Different Problems
What customers say — “we want to move data back into ERP/CRM” — technically and commercially blends four completely different things. Pulling them apart is the foundation of every judgment in this report, and the most incisive questioning framework for a Discovery conversation.
| Sub-problem | What it really is | Who solves it today | Status judgment |
| D1 Data Sync-Back | Sync lakehouse-computed fields (scores, segments, labels, recommendations) onto fields of business-system objects | Reverse ETL (Hightouch, Fivetran Activations); zero-copy sharing (Salesforce, SAP, ServiceNow) | Solved, commoditizing — packaged by M&A, squeezed by zero-copy |
| D2 Transactional Writeback | Create business-semantic “documents” inside the business system: invoices, sales orders, journal entries, work orders — must pass through the ERP’s internal machinery of validation, master data, authorization objects, posting periods | Almost nobody: Fivetran Activations is deep only on NetSuite (34 objects); Hightouch reaches only Sales Orders on S/4HANA; everything else = custom integration / iPaaS project | Half-solved, thinnest in market — supply is nearly empty |
| D3 Process Execution | Multi-step, cross-system business processes with approvals, exception handling, and compensation, triggered by an insight or event and running recurrently over time | BPM / RPA / agent platforms — but each executes only inside its own platform island | Unsolved in a cross-vendor context |
| D4 Evidence & Audit | Who authorized it, what was changed, on what insight, producing what business outcome — immutable, exportable, acceptable to audit and compliance | Nobody. Every platform has run logs only (mutable, retention-limited, decoupled from business meaning) | Zero supply |
The four sub-problems have different buyers (D1 sells to data teams; D2 and D3 to business and IT operations; D4 to the CFO, compliance, and audit), different pricing logic (D1 by data volume, D2/D3 by process value, D4 by risk), and completely different competitive dynamics. The market has treated D1 as the whole problem — while the pain customers actually pay for is in D2, D3, and D4. This is exactly why partners like Blueprint keep asking about workflow, integration, and orchestration: their customers are stuck in D2–D4, while the tools in their hands cover only D1.
Judgment: any vendor who understands “writeback” as reverse ETL only is in the D1 red ocean; any vendor who can package D2–D4 into a single repeatable, re-purchasable product has no direct competitor today.
2. The Status Quo: How Customers Actually Do It Today (Eight Practices and Their Break Points)
In a world without an execution layer, what happens after “Great. Now what?” takes only eight forms today. Each has a real reason to exist and a structural break point — and the break point is Metaprise’s entry point.
| Practice | Typical form | Coverage | Break point |
| 1. Manual re-keying (swivel chair) | Analyst reads the dashboard then hand-enters; CSV bulk upload (Salesforce Data Loader, SAP migration tools) | D1–D2 (low volume) | Slow, error-prone, no evidence, non-reusable. MIT’s research names “tools not entering the workflow” as the core mechanism behind the 95% pilot-failure figure — this is the most common status quo |
| 2. Custom pipelines | Databricks Jobs / Airflow scripts calling Bulk API, OData, BAPI directly | D1–D2 | No compensation on partial failure (500 of 2,000 postings succeed then it aborts; re-runs risk duplicates), idempotency hand-coded, maintenance unowned, approvals over email; rebuilt from scratch on every project |
| 3. Reverse ETL | Hightouch / Fivetran Activations sync Delta-table fields to SaaS objects | D1; touches the edge of D2 on NetSuite | Field sync ≠ process; no per-record approvals; ERP coverage extremely narrow (§3.1); audit = run logs; no rollback |
| 4. iPaaS | Workato / Boomi / MuleSoft / Power Automate: job completes → webhook → recipe maps → API calls | D1–D2, some D3 | Shallow process semantics (approval is just a step), no compensation semantics, per-task pricing punishes high-frequency execution, logs unusable as audit evidence (§3.3) |
| 5. RPA | UI-level robots type results into the ERP screen | D2 (brittle) | Breaks the moment the UI changes; vendors are themselves rebuilding it with agentic (UiPath Maestro) |
| 6. Zero-copy + single-vendor execution | Data Cloud share → Data Actions → Flow; SAP BDC → Joule; ServiceNow WDF → Now workflows | D3 inside a single island | Execution endpoints all land in the counterparty’s platform and licensing; cross-system processes break; Salesforce federated live-query data cannot trigger Data Actions (§3.2) |
| 7. BPM platforms | Build process apps on ServiceNow / Pega / Appian / Camunda | D3 (in-platform) + real approvals | Heavy to build, insight not native, strong platform gravity; evidence is still logs; no outcome pricing |
| 8. Agent + MCP (emerging) | Agent reads the lakehouse via MCP, calls business-system APIs | D2–D3 in theory | Governance is call-level not process-level; no business-object semantics, idempotency, or compensation; production deployments rare (Camunda: 11% in production, 80% Q&A) |
Note two cross-cutting facts. First, approval is “borrowed” across all eight practices — borrowed email, borrowed Slack, borrowed Teams cards. No practice models “who is authorized to approve this write” as a first-class citizen, and nobody turns the approval record into audit evidence. Second, no practice is safe when it “fails halfway”: after a bulk write aborts, do you re-run the remainder or not? Without idempotency and compensation semantics, that question has no good answer — which is the technical root of why enterprises dare not let AI touch systems of record, and one of the most concrete reasons behind Gartner’s “40% of agentic projects will be cancelled.”
3. The Full Market Landscape: A Six-Layer Vendor Map
The following unfolds six layers in order from “closest to the data” to “closest to the process”: Reverse ETL & Data Activation (§3.1), Databricks native & zero-copy alliances (§3.2), iPaaS & embedded integration (§3.3), the process execution layer — BPM, RPA, and the giants’ agent platforms (§3.4), execution infrastructure & fragmented startups (§3.5), and the structural conclusions that cut across all layers (§4).
3.1 Reverse ETL / Data Activation: The Last Mile of “Data,” Not the Last Mile of “Process”
Reverse ETL was the market’s first-generation answer to “writeback,” and today it is the only productized one. Its 2025–2026 story is one of a category disappearing: independent vendors are either being acquired into data-pipeline platforms or escaping up-stack into marketing-domain AI decisioning.
| Vendor | Status as of July 2026 | ERP writeback capability | Governance (approval / audit / rollback) |
| Hightouch | Independent survivor: closed a $150M Series D at a $2.75B valuation in Apr 2026; leads with marketing-domain AI Decisioning; deepest Databricks relationship (Partner Connect listing, Databricks Ventures investee, 2025 Retail ISV Partner of the Year) | NetSuite (behind the Business-tier paywall, can write invoices / sales orders / JEs); S/4HANA: only the single “Sales Order” object (OData); Workday, SAP IBP limited; no Oracle Fusion, no Dynamics F&O | Approval Flows govern configuration changes only, not per-record writes; audit = sync run logs; no rollback |
| Fivetran Activations (formerly Census) | Acquired by Fivetran May 2025; Fivetran completed its merger with dbt Labs on 2026-06-01 — writeback is now one module of a pipeline platform (billed by MAR consumption) | The deepest NetSuite write in the category: 34 objects (invoices, POs, JEs, bills, etc.), ~500 records/min; no S/4, Fusion, Workday, or F&O writes | No per-record approval; no rollback; log-level audit |
| RudderStack | Remains independent; pivoted to “customer context engine for AI” in Feb 2026 and released IaC governance | No ERP writes (martech / CRM oriented) | Pipeline-config governance, not write governance |
| Twilio Segment | Linked Audiences reads the warehouse directly (incl. Databricks), activates to ~450 martech destinations | No ERP | Same as above |
| GrowthLoop / MessageGears / DinMo | All bet on agentic marketing; MessageGears announced a Databricks partnership in May 2025 | No ERP | Same as above |
| Polytomic / Skyvia | Engineering-oriented bidirectional sync / SMB long-tail tool | Polytomic supports NetSuite; Skyvia shallow multi-connector | RBAC + logs |
| CDP side: Amperity, ActionIQ (into Uniphore), Simon AI | Amperity went lakehouse-native and launched the Chuck Data agent; ActionIQ acquired by Uniphore; Simon Data renamed Simon AI, betting on Snowflake | No ERP | Same as above |
The way customers actually use it is highly uniform: a gold table (segments, scores, next-best-action) → diffed on a ~15-minute schedule → upsert the fields onto one SaaS object. That is where it ends. The field “churn risk = high” lands on the Account, but the renewal-defense play — create the opportunity, adjust the discount, open the ticket, notify the CSM, leave evidence — is entirely out of product scope.
Category judgment: D1 is commoditized — packaged by M&A (Fivetran / dbt), squeezed by zero-copy, absorbed from the lakehouse side by Lakebase, and its survivors are fleeing collectively toward marketing decisioning. Metaprise should not compete in this layer; it should treat it as the input end: everything after the insight lands in a Delta table is where our product begins. And remember the counterintuitive fact: the entire category’s write capability into SAP S/4HANA is a single object (Sales Order) — the supply side of D2 is essentially empty.
3.2 Databricks Native Capabilities and Zero-Copy Alliances: A Closed Loop on the Read Side, Islands on Execution
To answer your core question — “Databricks solved data integration and governance, but customers always want to move data back into ERP/CRM; how is that done today?” — we first have to see clearly how far Databricks itself has gone. The conclusion: the read side and the agent-building side are already a closed loop; the write side and the execution side are deliberately left blank. Everything below is verified 2025–2026 fact.
- Lakebase (GA: 2026-02 AWS / 2026-03 Azure): managed Postgres on the lakehouse. Its official positioning is to “replace the reverse-ETL pipelines that sync data back to operational databases” — but the audience it serves is net-new apps built on Databricks. What it eliminates is the “lakehouse → your own app” return flow; it does not write to SAP and does not write to Salesforce.
- Agent Bricks (GA) + MCP Services (Beta): agents can call external APIs via MCP services registered in Unity Catalog, with per-call allow / deny / require-approval policies, and call records logged to a system table. But the built-in services are only Slack, GitHub, Atlassian, Google Workspace, and SharePoint — no managed action connector for any system of record (SAP, Salesforce, ServiceNow, Workday, NetSuite, Dynamics); approval is a “call-level gate,” not a business approval flow with separation of duties (SoD); no idempotency, compensation, or cross-system evidence chain.
- Unity AI Gateway (DAIS 2026): a single control point for agent / model / tool traffic (quotas, content filtering, MCP registry, audit) — it governs “access,” not “process outcomes.”
- Genie (formerly Databricks One, GA 2026-01) + Genie Agents: the business-user entry point; the Genie Conversation API is GA (2026-04); demos already show directional signals of “update records in Workday” via MCP — note, a signal, not a product.
- Lakeflow: 100+ connectors are all “in,” zero “out.” Neither DAIS 2025 nor 2026 shipped any writeback / activation product; Partner Connect maintains a dedicated reverse-ETL category (only Hightouch, Census), and Databricks Ventures chose to invest in Hightouch rather than build — writeback is institutionally outsourced to the ecosystem.
- One line of business context: valued at $134B (2025-12, $4.8B run-rate, +55% YoY), with a strong 2026 IPO expectation. It has plenty of ammunition — but all of it is currently aimed at closing the loop inside the lakehouse.
READ / WRITE / EXECUTE Semantics of the Five Major Alliances
| Alliance | READ (maturity) | WRITE (true semantics) | EXECUTE (execution endpoint) |
| Salesforce | Data 360 ↔ Databricks bidirectional zero-copy, GA; expansion announced 2026-06-16 (“trusted data into trusted action”) | Result tables can be shared into Data 360; federated live-query data cannot trigger Data Actions — event triggering requires a materialized / cached copy first | Flow / Agentforce, inside the Salesforce domain only; the Informatica acquisition closed 2025-11-18, internalizing the data pipeline |
| SAP | Business Data Cloud + SAP Databricks GA (2025-04); BDC Connect for Databricks GA (2025-10), bidirectional zero-copy | The return flow reaches only SAP’s analytics / planning layer (SAC, Joule context) — never S/4 transactions; postings still go through standard BTP interfaces | Joule agents + SAP Build (inside the SAP domain); SAP made a strategic investment in n8n in 2026-05 (valued $5.2B) and embedded it into Joule Studio — SAP is building its own cheap execution fabric |
| ServiceNow | Workflow Data Fabric zero-copy, bidirectional; the 2025-05 Workflow Data Network lists Databricks as a launch partner | Lakehouse insights can directly trigger Now workflows (an explicit design goal) | The most complete, but all inside the Now platform and its licensing; Knowledge 2026: Action Fabric + Autonomous Workforce + AI Control Tower expansion; the Moveworks acquisition closed 2025-12-15 |
| Palantir | Strategic partnership 2025-03; thin public product detail (a “100+ joint customers” blog) | Foundry Ontology Actions: webhook side effects write back to source systems, including documented user-level SAP OAuth writeback | Closed loop inside AIP — proprietary, services-heavy, ontology lock-in; it is the reference model for the “ontology + action” loop, not an open layer |
| Microsoft | Azure Databricks ↔ Power Platform connector (2025-06 preview): Power Apps can CRUD Databricks tables | Power Automate can execute SQL against Databricks, then reach business systems via 1,400+ connectors | Today the most concrete “lakehouse → business workflow” bridge — but governance and approval all belong to the Power platform, the Microsoft island; coexists in coopetition with Fabric |
Key conclusion — as of mid-2026, the six things Databricks + all its alliances still do NOT do: (1) no native writeback / activation product, Lakeflow only ingests; (2) agent actions cannot reach systems of record, built-in MCP services are collaboration apps only; (3) governance stops at the call level — no multi-level business approval chain with SoD, no cross-system transaction semantics (saga / compensation), no rollback of external side effects — Lakebase branching only rolls back state Databricks itself manages; (4) Jobs orchestrates compute DAGs, not a long-running “human + system” process state machine, and there is no recurring-mission abstraction spanning SAP → Salesforce → ServiceNow; (5) every alliance’s execution endpoint is a single-vendor island — cross-system, one approval model, one evidence chain, one rollback story, nobody owns it; (6) no commercial container for “executable business outcomes”: no outcome pricing (the whole line is DBU consumption), Marketplace Apps monetization is still in pilot, partner process IP has nowhere to be monetized.
The threat clock: MCP Services’ require-approval policy, Unity AI Gateway, and the Genie Agents Workday demo show that Databricks is growing action-layer primitives from the inside out. Reasonably expect managed action connectors and a more complete approval UX within 12–18 months. The window for an independent execution layer is real but narrowing — the correct posture is not to fight it but to plug in along its rails (§5.2), building differentiation on “multi-system process semantics + evidence + economic model,” not on “can call an API.”
3.3 iPaaS and Embedded Integration: The Plumbing Is Becoming “Agent Infrastructure,” but Nobody Owns the “Outcome”
iPaaS is today the most common serious answer customers use for “lakehouse → business system,” and the toolbox the Blueprints of the world know best. In 2025–2026 the whole category rewrote its narrative: Gartner’s March 2026 iPaaS Magic Quadrant reframed the category around AI workflows, agent orchestration, and AI gateways, and defined MCP as “the vendor-neutral bridge between the AI world and third-party enterprise systems.” Every vendor shipped an agent builder and an agent governance console — but not one touches evidence-grade process governance, rollback semantics, outcome pricing, or partner revenue share.
| Vendor | Key 2025–26 moves | Strength | Gap on D2–D4 |
| Workato | Workato ONE + Genies (2025); AIRO trust layer (actions attributed to the initiating human, not a service account); AI Marketplace (2026, no public rev-share terms) | Broad connectors, deep enterprise penetration, strongest agentic narrative (furthest in Gartner 2026 Vision) | Approval is just a step in a recipe; logs are not evidence; per-task billing punishes high frequency; SIs get no product rev share |
| Boomi | Agentstudio (2025-01) + Agent Control Tower (register / observe third-party agents); LOI to acquire MCP gateway Lunar.dev in 2026-05 | Best third-party agent registry and observability | Governance = observability, not process semantics; no rollback |
| MuleSoft (Salesforce) | Agent Fabric GA (2025-10): Agent Registry, Broker, Flex Gateway (MCP + A2A), Visualizer | API-led deep integration; the connectivity substrate for Agentforce | Approval / audit delegated to the Salesforce platform; heavy vCore pricing (typical enterprise entry $250k+/yr); in-domain gravity |
| SnapLogic / Celigo / Jitterbit | AgentCreator 3.0 (HITL upgrade) / Ora Agent Builder (2026-03) / MCP repositioning manifesto | Mid-market coverage; Celigo’s O2C process templates are closest to process semantics | Same category ailment: evidence, compensation, outcome all missing |
| IBM (webMethods + watsonx Orchestrate) | Relaunched Hybrid Integration in 2025; Orchestrate does agent orchestration, partnered with Oracle OCI in 2026 | Hybrid / on-prem ERP legacy scenarios | Two-product seam; license pricing; innovation pace |
| n8n | $180M Series C @ $2.5B in 2025-10 → SAP strategic investment @ $5.2B in 2026-05, embedded into Joule Studio; self-hosting + per-execution billing | Cheap execution fabric, strongest developer momentum; billed per execution, not per step | Thin RBAC / audit; DIY governance; becoming an asset inside SAP accounts |
| Zapier / Make | Zapier Agents + MCP; Make (Celonis-owned) AI Agents | Long-tail ubiquity | Team-level ceiling; light governance |
| Power Automate / Logic Apps | Native Approvals (Teams / Outlook cards, the best human gate in the ecosystem); agent flows + Logic Apps agent loop; dedicated Databricks connector (2025-06) | Approval experience + Purview compliance ecosystem + distribution | Activity logs ≠ outcome evidence; license complexity; the Microsoft island |
| SAP Integration Suite / Oracle OIC | The “front door” for ERP writeback: CPI iFlow + OData / BAPI / IDoc; OIC + FBDI; Oracle Fusion Agentic Applications (2026-03) | The official channel and compliance stance for writeback | In-vendor domain; zero cross-system neutrality; heavy to build |
| Embedded integration layer | Nango, Composio ($25M, 2025-07), Arcade.dev ($60M, 2026-06, agent action-auth layer), Paragon, Merge; Pipedream acquired by Workday in 2025-11 | Our OEM options for expanding connector breadth | Access layer only, no process / evidence; watch out for being acquired (the Pipedream lesson) |
On the ERP last mile, one point is worth spelling out: to write a single document into S/4HANA, you pass through released OData / BAPI interfaces, master-data validation (cost center, account, material), number ranges, posting-period checks, and authorization objects — these mechanisms live inside the ERP, the middleware can’t see them, and only receives an ABAP error when a call fails. Writing to Oracle Fusion is the same (REST + FBDI import jobs). This is why iPaaS projects on ERP writeback forever “insert records around the transaction machine,” and why approvals, exception queues, and evidence carry the highest premium at this layer.
iPaaS’s four “structural cannots”: cannot do process — the engine is a stateless step-executor, approval is just a step, compensation is hand-coded, and the long-running “human + system” state machine does not exist; cannot do evidence — operational telemetry is mutable, retention-limited, and decoupled from business meaning; audit rejects it; cannot do outcome pricing — per-task / per-connection / vCore billing is decoupled from value, and switching to outcome billing cannibalizes the revenue statement; cannot turn SI IP into product — every marketplace (Workato, Anypoint Exchange, Boomi, AppSource) is a lead-gen catalog where partner assets are free accelerators traded for services opportunities; the platform does not meter partner IP, so there is no unit to share. The fourth point matters most to us: iPaaS physically lacks a “meterable unit that can be shared” — and the Mission is precisely that unit.
3.4 The Process Execution Layer: BPM, RPA-Turned-Agentic, and the Giants’ Platforms
This layer is the true protagonist of “execution” and the core of the competitive analysis. The overall picture: whoever does governed execution most completely is locked inside their own platform gravity; whoever is cross-system neutral has no governance; and anyone who combines both plus an outcome economic model does not exist.
| Player | Positioning and key 2025–26 moves | Closest point to us | Structural gap |
| ServiceNow | Knowledge 2026, verbatim: “Others let agents read and write data. We let agents execute governed work”; Action Fabric (exposes Now workflows / approval chains to external agents, with Anthropic as launch partner) + Autonomous Workforce + AI Control Tower expansion; zero-copy direct-connect to Databricks; the Store has real partner rev share | The most dangerous direct competitor — the complete version of governed execution, and it already said our sentence | Everything inside the Now platform and its licensing (not neutral, not portable); president Zavery publicly rejects outcome pricing (“Nobody’s been able to make it work”); no compensation semantics; evidence not portable |
| Pega | Agentic Process Fabric (2025-06); Infinity ’26 exposes processes externally via MCP; charges a flat fee per completed case | The market’s closest precedent to per-mission outcome pricing | Small ecosystem, weak partner economics, case-platform gravity |
| Appian / Camunda | Appian Agent Studio (25.4); Camunda 8.8 agentic BPMN orchestration — the only player with true BPMN compensation semantics; its 2026 report is the best data source on adoption reality | The textbook of process semantics | Sells an engine to engineering orgs; no connector depth, no evidence product, no outcome |
| UiPath | Maestro (BPMN + agents + robots + HITL); 2025.10: multi-vendor agent orchestration (MCP) + Data Fabric preview: zero-copy Databricks / Snowflake triggering processes | The product shape closest to Metaprise | IT-automation buyer and brand; audit not evidence-grade; growth under pressure (ARR +12%); no outcome |
| Automation Anywhere | EnterpriseClaw: universal orchestration across Salesforce / ServiceNow / SAP (preview, 2026) | Same as above, one notch weaker | Scale and mindshare weaker than UiPath |
| Microsoft | Agent 365 GA (2026-05-01): agent registry, Entra Agent ID, Purview audit; Copilot Studio multi-agent + computer use | Distribution is unbeatable; “good enough by default” is how it threatens | M365 gravity; no business process / case semantics; two control planes coexist (Foundry vs Agent 365) |
| Salesforce | Agentforce 3 / 360; AgentExchange (partner rev-share precedent); Flex Credits (~$0.10 / action) is “quasi-outcome” pricing; ~9,500 paid deals / $1.4B ARR in 2025 (bundled-with-Data-360 basis) | Partner rev share + the largest agent revenue base | Inside the CRM domain — “executes your Salesforce, not your enterprise”; trust and pricing confusion persist |
| AWS / Google / OpenAI | Bedrock AgentCore GA + Policy GA (2026-03, Cedar pre-execution policy gate); Gemini Enterprise + A2A (donated to the Linux Foundation, 150+ orgs); OpenAI AgentKit + Frontier (2026-02, enterprise agent-management platform) | Commoditizing “agent management” from the bottom | Developer infrastructure; no business approval UX, no process semantics, no partner-IP economics |
| Celonis | The process-intelligence graph is inherently “continuous sensing”; the Orchestration Engine (with Emporix) grows cross-system execution; AgentC feeds process context to other vendors’ agents | Owns the front door of “insight triggering” | Execution is young and OEM’d; shallow approval-authority model; consulting-style monetization; no evidence product |
| Temporal et al. durable infrastructure | Temporal $300M Series D @ $5B (2026-02, “to make agentic AI real”); Orkes, Restate, Inngest, DBOS | Substrate options for our runtime / DIY competition | Building blocks, not products; approvals, audit, and billing all left for the customer to write |
| Fragmented startups | Arcade.dev $60M (action auth); Kognitos $25M (deterministic English-defined processes, ERP-oriented); HumanLayer / gotoHuman (approval APIs); Lyzr, Beam, Tektonic; Rubrik Agent Rewind (2025-08) turned “rolling back AI actions” into a named market need | Every single point has someone doing it | Nobody has the combination; mostly acquisition targets rather than category owners |
Research conclusion (verbatim-level): no VC-backed or public company is selling the complete bundle — insight-triggered + graded human authority + immutable evidence + compensation/rollback + outcome-based pricing + a partner-IP revenue-share marketplace. Every fragment exists; the combination does not. Two of these are zero-supply market-wide: turning the evidence chain into a product (everyone has logs only), and billing by mission outcome (ServiceNow’s president publicly says it can’t be done; Pega’s per-completed-case fee is the only adjacent move). These two sentences should appear in every one of our Partnership Briefs.
4. Structural Conclusions: Why This Gap Remains Unclaimed, and Why Now Is the Timing
4.1 Three DNA Reasons Nobody Has Claimed It
Architecture DNA. Every existing engine class has a data model that was not built for “cross-system governed execution”: reverse ETL is a row-diff synchronizer; iPaaS is a stateless step-executor; BPM is a single-platform case system; RPA is a UI driver; the agent runtime is a reasoning loop. They all lack the same set of first-class citizens: cross-system long-running state, graded authority (SoD), a compensation graph, and immutable evidence. “Adding” these onto any one of them means rewriting the runtime.
Business-model DNA. A vendor billing by task / connection / seat / vCore grows revenue with “volume” while decoupling it from “value” — switching to outcome billing immediately cannibalizes its own revenue statement, which is the real context behind ServiceNow’s president publicly rejecting outcome pricing. Likewise, their marketplaces are all lead-gen catalogs: a platform that does not meter the execution of partner IP can never compute “how much the partner should be paid.” The billing unit determines the ceiling of the business model — the Mission unit is itself the moat.
Buyer DNA. Integration tools sell to integration teams, BPM to IT, agent platforms to innovation departments — while the buyer for D4 (evidence) is the CFO, compliance, and audit, and the budget for D2/D3 hangs on business outcomes. The incumbents’ sales motion cannot reach the people who pay for execution and evidence.
4.2 The Timing Argument: Moving Data Is Devaluing, Governing Execution Is Appreciating
SAP Databricks GA in 2025-04, ServiceNow Workflow Data Network in 2025-05, SAP BDC Connect GA in 2025-10, the Salesforce alliance expansion in 2026-06 — zero-copy sharing became the default architecture within a year. Sharing replaces moving, and D1 goes to zero; but zero-copy moves “data visibility,” not “actions” — it kills the sync business while exposing the execution gap even more starkly. At the same time, agents drive the marginal cost of “producing insight” toward zero, insight output explodes, and the bottleneck on the execution side is amplified in step. This is the evidence version of the “sixth layer” narrative: intelligence generation is solved and commoditized; intelligence execution is the next enterprise-grade challenge — and right now nobody owns it.
4.3 The Competitive Clock: Who Is Closing In
- ServiceNow (fastest): Action Fabric is expected to reach GA in H2 2026 — the day it exposes “governed execution” to external agents is the day it collides head-on with our narrative. Differentiation must land on: cross-platform neutrality, outcome pricing, portable evidence, and partner-IP economics.
- Databricks (12–18 months): the distance from MCP Services in Beta to managed action connectors. The countermeasure is not to sprint ahead but to register Metaprise as an “action tool” on its governance rails (§5.2), turning its evolution into our distribution.
- UiPath: if Data Fabric (zero-copy Databricks triggering) reaches GA and stacks evidence / outcome on top, it collides head-on — constrained by its growth pressure and IT-buyer gravity, medium probability.
- Celonis: growing “execution” from the “sensing” end; if it adds authority and evidence, it becomes Metaprise grown from the other side; its execution muscle is still OEM’d today.
- Microsoft / OpenAI (from below): Agent 365 and Frontier will commoditize “agent registration / observation / access governance” — which actually raises the relative value of our layer (process semantics + evidence + economic model), but demands we never let the positioning slide down to “agent management.”
4.4 Adoption Reality: Hard Data on the Execution Gap
| Data point | Source and date |
| 95% of enterprise GenAI pilots show no measurable P&L impact; the core mechanism is “tools not entering the workflow” | MIT NANDA, State of AI in Business 2025, 2025-08 |
| 40%+ of agentic AI projects will be cancelled by end-2027 (cost, unclear value, inadequate risk controls); of thousands of “agentic” vendors, only ~130 are the real thing | Gartner press release, 2025-06-25 |
| Only 17% of organizations have truly deployed agents (60%+ plan to within two years); agentic AI is crossing the peak of expectations into the trough | Gartner Agentic AI Hype Cycle, 2026 |
| 71% of enterprises are “using” agents, but only 11% of initiatives reach production; 80% of deployed agents are Q&A / chat rather than operators; 66% cite compliance as a barrier; 84% fear the business risk of insufficient oversight | Camunda, State of Agentic Orchestration, 2026-01 |
| 42% of companies abandoned most of their AI initiatives in 2025 (up from 17% the prior year); ~46% of PoCs die before production | S&P Global 451 Research, 2025-03 |
| 88% of organizations use AI regularly, but only 39% report enterprise-level EBIT impact; only 23% are scaling agentic AI | McKinsey, State of AI, 2025-11 |
| Only 25% of AI initiatives achieved expected ROI (survey of 2,000 CEOs) | IBM IBV CEO Study, 2025-05 |
| Only 25% of organizations moved 40%+ of pilots into production; among agent adopters, only 21% have mature governance | Deloitte, State of AI in the Enterprise 2026, 2026-01 |
| Demand-side counterpoint: CEOs plan to double AI spend in 2026 (~1.7% of revenue); 30%+ of AI budgets flow to agentic; 90% of CEOs expect measurable agent returns in 2026 | BCG AI Radar 2026, 2026-01 |
How to read it: the supply side (first eight rows) proves “execution is the bottleneck”; the demand side (last row) proves “budgets are pouring toward the bottleneck.” Pain and money coexist, and both are accelerating — this is the complete proof that the entry timing holds.
5. What Metaprise Can Contribute to This Pain Point
Using the four sub-problems of §1 as a coordinate system, Metaprise’s contribution can be located precisely — where not to play, where to attack, and where the moat lies that others cannot copy.
5.1 Contribution Mapped Against D1–D4
- D1 (don’t build, plug in): don’t compete with Hightouch / Fivetran on field sync; cede marketing-domain activation to them. Subscribe to “insight ready” events via a Delta Sharing recipient + Change Data Feed, or via a Lakeflow table update trigger, treating D1 as the input end. Our product begins “after the field lands in the table.”
- D2 (attack): make transactional writeback a first-class step of a Mission — business-object semantics (invoice / PO / JE / work order, not “a row”), idempotency keys, sandbox and test-mode, master-data pre-validation, failures routed to an exception queue rather than a log. Supply-side evidence: the entire reverse-ETL category’s S/4 write capability = 1 object; Fusion / Workday / F&O = 0. This is the thinnest layer in the market, the most painful for customers, and the most strictly audited — and therefore the layer where approval and evidence carry the highest premium.
- D3 (product core): a Mission = an insight-triggered, long-running human-machine state machine — CSE continuous sensing → multi-agent orchestration (nine patterns) → L1–L4 graded approval (with SoD) → cross-system execution → compensation / rollback → recurring runs and continuous improvement. In the market, only Camunda has true compensation semantics and only BPM has true approvals — nobody combines these two with “insight-triggered + cross-vendor neutral.”
- D4 (the king of differentiation): upgrade AuditChain from a feature to a product — the Execution Receipt: every Mission produces an immutable evidence package (the triggering insight and its data snapshot, the approver and authority level, the before/after image of every system write, the business-outcome metric), exportable in one click mapped to SOX / HIPAA / FFIEC / EU AI Act categories. Zero supply market-wide; Camunda’s “66% cite compliance as a barrier” is its demand-side proof. This is also the only way the team’s decade of financial-infrastructure DNA can be monetized that cannot be quickly imitated.
5.2 Seven Technical Integration Points with the Databricks Ecosystem (All Available Today)
- Lakeflow table update triggers (GA): a customer gold table updates → a job → a webhook calls the Metaprise Mission API — the lightest “insight ready” push, officially supported.
- Delta Sharing recipient + CDF: Metaprise, as a share recipient, read-subscribes to result tables and change data — no customer compute, cross-cloud, open protocol, inherently neutral.
- Genie Conversation API (GA 2026-04): embed governed natural-language querying inside the Mission approval interface — the approver can ask on the spot “where did this number come from,” turning approval from a blind sign-off into an evidenced decision.
- Model Serving / Agent endpoints: mid-Mission, call back to the customer’s own model / agent for a second judgment — the customer’s AI investment appreciates inside our process.
- Publish a Metaprise MCP Server, registered into the customer’s Unity Catalog MCP Services: let Databricks agents call Metaprise as a “governed action tool with an approval gate.” Ride its call-level approval rails and connect process-level governance on top — its agent proposes, our Mission executes. This step converts the threat of “will Databricks build it themselves” into distribution.
- Marketplace Apps (public preview 2026-06) + Partner Connect listing: enter its distribution surface as an App; simultaneously push to establish an “Execution / Agent Orchestration” partner category — ample precedent: the reverse-ETL category itself is the institutionalized proof that Databricks outsources writeback to partners.
- ZeroBus telemetry writeback: write a Mission’s execution results and evidence summary back into the customer’s lakehouse, so the customer sees “what was executed and what it was worth” in their own BI — the evidence loop returns to Databricks and the alliance narrative is complete.
5.3 The Role of the AI Product Line: Agents Propose, Missions Dispose
The implication for future AI products can be compressed into one product philosophy: agents propose, missions dispose. The reality of 2026 is that agents can already “think” — Databricks, Microsoft, and customers themselves can all build agents — but enterprises dare not let them “act”: Camunda’s 84% (fear of loss of control) and 66% (compliance barriers), and Gartner’s 40% (will be cancelled) are all saying the same thing. Metaprise’s AI product does not need to compete with Agent Bricks, Copilot, or Agentforce on “who is smarter” — it becomes the governance shell for all agents: an agent proposal from any source (change price, restock, open a ticket, adjust a credit line) enters a Mission, is executed with graded approval and compensation, and produces a receipt. The more prosperous and commoditized the model and agent layers become, the more valuable the execution shell — which is precisely the definition of the “thin waist” position.
5.4 The Economic Model Lands Exactly on the Two Zero-Supply Points
The two zero-supply points the research confirmed — outcome-based billing and an evidence-grade audit chain — sit exactly where L5 (Settlement) and L3 (AuditChain) live in the Metaprise stack. Billing per completed Mission lets us say to partners what iPaaS physically cannot: “every time your IP is executed, you get paid” (contrast: Workato’s marketplace has no rev share, ServiceNow’s Store has rev share but locked to the Now platform, AgentExchange takes ~15% but only inside the Salesforce domain). Pega’s per-completed-case fee proves enterprises accept this pricing — but Pega has no cross-platform neutrality and no partner marketplace. The promise of a second revenue curve is only credible when a “meterable execution unit + a public revenue-share structure” exist simultaneously — and we are the only ones who have both.
6. What We Still Lack: Capability Gaps and Priorities
The following is ordered P0 (next quarter) / P1 (within two quarters) / P2 (ongoing). The principle: attack the thinnest part of the market (D2), lock the window (Databricks integration), build the moat (D4 evidence).
| Level | Gap | Why, and what it looks like |
| P0 | Transactional ERP-connector depth | Start with 3–5 high-frequency documents for NetSuite + S/4HANA (invoice, sales order, JE, credit memo), then expand to Fusion / Workday / F&O; each connector carries idempotency keys, compensating actions, sandbox mode, and master-data pre-validation. Connector strategy: build deep connectors for the first 6–8 systems + use MCP for the long tail + OEM Nango / Composio / Arcade for breadth where needed. Cautionary note: Pipedream has already been acquired by Workday — breadth can be rented, depth must be owned. |
| P0 | Databricks-native integration + listing | Land the seven integration points of §5.2; list on Partner Connect / Marketplace Apps; build the first joint reference implementation with Blueprint. The 12–18-month threat clock sets this item’s priority — the “ride the rails” posture converts Databricks’ action-layer evolution from a death sentence into distribution. |
| P0 | Productize the Execution Receipt evidence | Define the audit export format (SOX / HIPAA / FFIEC / EU AI Act category mapping), an interface for CFOs and auditors, and end every demo by “presenting the receipt.” Zero supply market-wide, and it requires governance DNA rather than feature development — this is the one differentiation nobody can copy within 12 months. |
| P1 | Make compensation / rollback explicit | Demonstrate “undo” in every demo; fix the sales question: “Last time a bulk write failed halfway, how did you handle it?” Rubrik Agent Rewind proves “rolling back AI actions” has become a named need, and Camunda is the only comparable semantics — turn it into our language. |
| P1 | Document the partner economics | Publish the Mission Store’s revenue-share structure and metering basis (partner share per Mission execution, marketplace take rate, runtime-revenue split). The comparison weapons are ready: Workato has no rev share, ServiceNow Store is platform-locked, AgentExchange’s 15% is the precedent. A “second revenue curve” must have signable terms to be credible. |
| P1 | Blueprint joint-offer template | One line: “The AI Factory ends at model production; the Execution Factory begins at model production.” Attach a 30-day lighthouse Mission: churn signal (Databricks) → Salesforce renewal ticket + NetSuite credit check + ServiceNow task, with L2 approval and an Execution Receipt. This is the first instance of the OTP (Outcome Transformation Partner) ICP, and the minimum verifiable proof of this report’s narrative. |
| P2 | Front-load compliance assets | A deliverable SOC 2 report, a FedRAMP-Ready narrative, and a data-residency / PII-handling statement on the front page of the sales kit. 66% of buyers cite compliance as a barrier — we treat it as a weapon. |
| P2 | Routinize competitive monitoring | Re-assess quarterly whether the §3.2 “six do-nots” list still holds; watch: ServiceNow Action Fabric GA (~H2 2026), Databricks managed action connectors, UiPath Data Fabric GA, SAP × n8n deepening, OpenAI Frontier enterprise penetration. |
7. Translating This into Partner-Brief Language
7.1 First-Page Narrative (Add Evidence to Your Version)
The AI market has largely solved intelligence generation. The next enterprise challenge is intelligence execution. Our partners already know how to generate insight. We help them operationalize that insight across the enterprise — with graded human authority, immutable evidence, and rollback — transform reusable expertise into governed missions, and create recurring revenue without replacing their existing business.
The three facts that back it (one footnote line each in the brief is enough):
- The execution gap has hard data: 95% of pilots show no P&L impact (MIT); only 11% of agent projects reach production (Camunda); 40%+ will be cancelled (Gartner).
- Databricks’ own Partner Connect “writeback” category has only two vendors, both marketing-oriented; Databricks Ventures invested in one of them rather than building — the platform left the execution layer to the ecosystem.
- Every zero-copy alliance’s (Salesforce, SAP, ServiceNow) execution endpoint is inside a single vendor’s platform — a cross-system governed execution layer is owned by no one today.
7.2 One-Sentence Positioning
Metaprise is the execution layer between every lakehouse and every system of record: cross-vendor missions with graded human authority, immutable evidence, rollback, outcome-based pricing, and partner-owned IP economics.
7.3 Two Objections You Must Answer
“Won’t Databricks just build it themselves?” — the evidence is that it hasn’t and institutionally doesn’t: Lakeflow’s 100+ connectors are all “in,” zero “out”; writeback is outsourced to Partner Connect; it invested in Hightouch rather than building; two DAIS with no writeback product. What it is building is call-level action primitives, and we register as an action tool on its governance rails via MCP Services — the more it does, the better our distribution. At the same time, acknowledge the threat clock (12–18 months) — which is precisely the reason to urge partners to move now.
“Isn’t this just ServiceNow?” — ServiceNow is the best single platform for governed execution, but has three structural differences: execution and evidence are locked inside the Now platform and its licensing (not neutral, not portable); its president publicly rejects outcome pricing; and partner rev share exists only inside the Store ecosystem. We are cross-platform neutral, bill by outcome, export evidence, and let partner IP monetize on any customer’s stack. For customers already heavily invested in ServiceNow, Action Fabric is instead an integration target for us, not a death sentence.
7.4 Supplementary Discovery / Qualify Questions (D2–D4 Probes)
- After the last model or analytics output, who — with what tool — entered the result into ERP / CRM, and how long did it take? (exposes D2)
- If a bulk write half-succeeds and half-fails, what is the handling process today? (exposes the missing compensation)
- The last time audit or compliance asked you to prove “who approved a given AI / analytics-driven change,” what did you hand them? (exposes D4)
- Which processes are stuck today because “changing data in the system requires approval,” so nobody dares automate them? (exposes D3 and the authority model)
- If the execution layer were billed separately, how would you want to pay — per call, per process, or per completed business outcome? (tests outcome acceptance, against the Pega precedent)
7.5 ICP Language
Upgrade the ICP from “IP-bearing Data & AI Services Partner” to the Outcome Transformation Partner (OTP): in the customer’s transformation, the partner owns strategy + delivery + industry knowledge, Metaprise owns the execution layer, and together they deliver the business outcome. Four criteria: the customer is undergoing transformation / modernization; the partner stops at recommendation today (model production is the delivery endpoint); the partner holds reusable but un-productized delivery IP (accelerators, frameworks, prompts, scripts); and the partner is actively seeking a second revenue curve. Blueprint hits all four — it is not a special case, it is the first instance of this ICP.
Appendix A: Research Scope and Methodology Note
This report is based on four parallel deep-research streams completed on 2026-07-28: (1) the Reverse ETL & Data Activation market; (2) Databricks native capabilities, the DAIS 2025 / 2026 announcements, and the five major alliances; (3) iPaaS, ERP-native integration channels, and embedded integration; (4) the process execution layer (BPM, RPA, the giants’ agent platforms, durable execution, startups) and adoption-reality data. Key facts (M&A and closing dates, product GA status, pricing models, executive statements, funding valuations) were cross-verified against at least two independent sources; single-source or vendor-self-reported items are flagged in the text with wording such as “preview / demo / not independently verified.” All judgment conclusions (“six do-nots,” “four cannots,” “the combination does not exist”) are inferences from verified facts and should be re-reviewed quarterly (see §6, P2).
Appendix B: Primary Sources
Reverse ETL / Data Activation
Fivetran acquires Census (2025-05): fivetran.com/press/fivetran-signs-agreement-to-acquire-census · techcrunch.com/2025/05/01/fivetran-acquires-census
Fivetran and dbt Labs complete merger (2026-06-01): fivetran.com/press/fivetran-dbt-labs-complete-merger
Fivetran Activations NetSuite destination (34 objects): fivetran.com/docs/activations/destinations/available-destinations/netsuite
Hightouch $150M Series D @ $2.75B (2026-04): pymnts.com/news/investment-tracker/2026/hightouch-valued-at-2-75-billion · S/4HANA destination (Sales Orders only): hightouch.com/docs/destinations/sap-s4hana · Approval Flows (config approvals only): hightouch.com/docs/workspace-management/approval-flows
Databricks Partner Connect reverse-ETL category (two members only): docs.databricks.com/aws/en/partners/reverse-etl/hightouch and /gcp/en/partners/reverse-etl/census · Databricks Ventures invests in Hightouch: databricks.com/blog/activating-data-lakehouse-databricks-ventures-invests-hightouch
RudderStack IaC governance and RudderAI (2026-02): prnewswire.com (RudderStack release) · Segment Linked Audiences: twilio.com/en-us/press/releases/linked-audiences · MessageGears × Databricks (2025-05): businesswire.com · Uniphore acquires ActionIQ: businesswire.com (2024-12-05)
Databricks Native and Alliances
Lakebase GA (2026-02 AWS / 2026-03 Azure): community.databricks.com (Lakebase GA) · databricks.com/blog/azure-databricks-lakebase-generally-available · Lakebase positioned to replace reverse ETL: databricks.com/en/blog/reverse-etl-lakebase-activate-your-lakehouse-data-operational-analytics
MCP Services (Beta, built-in service list and require-approval policy): docs.databricks.com/aws/en/agents/agent-framework/mcp-services · External connection tools: docs.databricks.com/aws/en/generative-ai/agent-framework/external-connection-tools
DAIS 2026 recap (Unity AI Gateway, Genie Agents, Marketplace Apps): atlan.com (DAIS 2026 announcements) · flexera.com/blog/perspectives/databricks-data-ai-summit-2026 · Marketplace Apps public preview (2026-06-16): databricks.com/blog/announcing-apps-databricks-marketplace
Table update triggers (GA): docs.databricks.com/aws/en/jobs/trigger-table-update · Genie Conversation API GA (2026-04): docs.databricks.com/aws/en/ai-bi/release-notes/2026
Salesforce × Databricks expansion (2026-06-16): salesforce.com/news/stories/salesforce-databricks-shared-foundation-of-human-agent-work-announcement · zero-copy mechanics and Data Actions limitation: salesforce.com/partners/databricks · Informatica close (2025-11-18): salesforce.com/news/press-releases
SAP Databricks GA (2025-04) and BDC Connect GA (2025-10): databricks.com/blog (two GA blogs) · Joule Studio (2026-05): news.sap.com · SAP invests in n8n @ $5.2B (2026-05): tech.eu
ServiceNow: zero-copy (2024-10) and Workflow Data Network (2025-05): newsroom.servicenow.com · Knowledge 2026 (Action Fabric, Autonomous Workforce, AI Control Tower): newsroom.servicenow.com · constellationr.com · rejecting outcome pricing: diginomica.com (Q1 2026 earnings commentary) · Moveworks close (2025-12-15): moveworks.com
Palantir × Databricks (2025-03): databricks.com/company/newsroom · Foundry Actions and SAP OAuth writeback: palantir.com/docs/foundry/action-types/webhooks and /docs/foundry/sap/oauth2-writeback
Azure Databricks Power Platform connector (2025-06): databricks.com/blog · learn.microsoft.com/en-us/connectors/databricks · Databricks $134B and $4.8B run-rate (2025-12): databricks.com/company/newsroom
iPaaS and Embedded Integration
Gartner 2026 iPaaS MQ (2026-03, MCP framing): workato.com/the-connector/gartner-magic-quadrant-2026 · Workato ONE and Genies: siliconangle.com (2025-08-19) · AI Marketplace: workato.com
Boomi Agentstudio and Agent Control Tower: boomi.com · Boomi World 2026 and Lunar.dev LOI: erp.today · MuleSoft Agent Fabric GA (2025-10): salesforce.com/news/stories/mulesoft-agent-fabric-announcement
n8n $180M @ $2.5B (2025-10): blog.n8n.io/series-c · SAP investment @ $5.2B into Joule Studio (2026-05): tech.eu · SnapLogic AgentCreator 3.0, Celigo Ora (2026-03), IBM webMethods Hybrid Integration: respective official news pages
SAP writeback mechanics (OData / BAPI / clean core): community.sap.com · Oracle Fusion Agentic Applications (2026-03-24): oracle.com/news · Pipedream acquired by Workday (2025-11-19): newsroom.workday.com · Composio $25M (2025-07): siliconangle.com · Arcade.dev $60M (2026-06-15): businesswire.com
Process Execution Layer and Adoption Data
ServiceNow “execute governed work” and Action Fabric: reworked.co · diginomica.com (Knowledge 2026) · Pega per-case pricing and Infinity ’26: siliconangle.com (2026-06-08) · Camunda 8.8 agentic and compensation semantics: camunda.com
UiPath Maestro and Data Fabric preview (2025.10): uipath.com/blog · ir.uipath.com (FY27 Q1) · Automation Anywhere 2026 platform (EnterpriseClaw): automationanywhere.com
Microsoft Agent 365 GA (2026-05-01): microsoft.com/en-us/security/blog · Copilot Studio governance update (2026-04): microsoft.com · Salesforce Agentforce data (9,500 paid / $1.4B ARR): salesforceben.com · salesforce.com (FY26 Q3) · AgentExchange (2025-03): salesforce.com
AWS AgentCore GA (2025-10) and Policy GA (2026-03): aws.amazon.com/about-aws/whats-new · OpenAI AgentKit (2025-10) and Frontier (2026-02): openai.com · techcrunch.com · Google Gemini Enterprise and A2A (Linux Foundation, 150+ orgs): linuxfoundation.org
Celonis Orchestration Engine (with Emporix) and AgentC: celonis.com/news · diginomica.com · Temporal $300M @ $5B (2026-02): temporal.io/blog · LangChain $125M @ $1.25B (2025-10): langchain.com/blog
Kognitos $25M (2025-06): businesswire.com · Rubrik Agent Rewind (2025-08): reuters.com (syndicated) · Tines $125M @ $1.125B (2025-02): prnewswire.com