Dodge AI Turns ERP Memory Into Agent Infrastructure

Dodge AI is building enterprise maintenance agents around a less glamorous asset than a foundation model: the institutional memory buried inside a company's systems.

Every large ERP environment develops its own logic. Custom objects, configuration choices, old tickets, business rules, integrations, and undocumented workarounds determine why the software behaves differently inside one company than it does everywhere else. The person who remembers those decisions can often solve an incident faster than the person with the deepest generic product knowledge.

Dodge wants to turn that memory into infrastructure. Founded in 2025 by CEO Rebhav Bharadwaj and CTO Aditya Thakur, the San Francisco and Bengaluru company is developing an AI control plane for application maintenance across SAP, Salesforce, Microsoft Dynamics, Kinaxis, and Oracle JDE.

The company recently announced a $2.65M Seed round backed by Accel, Google AI Futures Fund, Schema Ventures, New Build Ventures, Antler, and SAP-ecosystem angels. The financing gives Dodge more room to prove whether enterprise context can support useful agents without weakening the controls that mission-critical systems require.

What Dodge AI Builds

Dodge's product starts by mapping an enterprise's application environment. Its public materials describe a context graph built from custom code, configurations, ticket history, documentation, and business processes. Retrieval models and orchestration agents use that graph to investigate incidents, answer support questions, and guide or execute changes.

That architecture matters because a generic model can know how SAP is supposed to work and still misunderstand a specific implementation. A field may feed a planning rule through a custom integration. A workaround may exist because of a decision made three years earlier. A technically valid change can still break the business process around it.

Dodge is trying to give agents the local explanation before they act. The company frames the system as a control plane rather than a single-purpose chatbot, with coverage across multiple enterprise platforms. That multi-system view is important because incidents rarely respect software boundaries. A visible failure in one application may begin with data, policy, or configuration somewhere else.

The Institutional-Memory Thesis

Enterprise application maintenance is often purchased as labor. Internal teams and outside consultants diagnose tickets, trace dependencies, document fixes, and carry knowledge from one change cycle to the next. The work is expensive partly because the same context must be reconstructed repeatedly.

Dodge's thesis is that the context itself can become a reusable software asset. Once the platform understands how a customer's systems fit together, each resolved incident can improve the record available for the next one. That creates the possibility of shorter support queues and faster changes without asking a fresh consultant to rediscover the environment every time.

It also creates a demanding product standard. The platform has to distinguish a repeatable pattern from a dangerous exception. It needs clear permissions, traceable recommendations, human escalation, and dependable rollback. In enterprise maintenance, a confident wrong answer can affect inventory, finance, fulfillment, or planning. The product earns trust through controlled behavior, not through fluent explanations.

Early Evidence and Its Limits

Dodge says it works with more than a dozen enterprises, half of them publicly listed, and processes hundreds of queries per hour. Those figures are company-reported and a complete customer list is not public.

The company's strongest public operating example is an anonymized CPG case study. Dodge reports 30% fewer MRP-related support tickets, 11 planner hours saved per plant each week, and four times faster solution delivery. Tech Funding News also relays a company-reported example in which an inventory-planning process became 132 times faster.

These results describe the value Dodge is pursuing, but they are not independent validation. The CPG customer is unnamed, and public sources do not disclose audited savings, contract values, retention, or a comparable deployment baseline. The next proof point is not a larger percentage. It is repeatability across named environments with different systems, controls, and failure modes.

Founders, Footprint, and Hiring

Rebhav Bharadwaj and Aditya Thakur met at BITS Pilani. Bharadwaj leads the company as CEO, while Thakur serves as CTO and previously co-founded Interface AI. Accel's official portfolio profile confirms the team and places Dodge inside the 2026 Accel Atoms and Google AI Futures Fund cohort.

The company operates across San Francisco and Bengaluru. A current founder post shows Dodge hiring product engineers in Bengaluru. There is no dedicated public careers page or reliable opening count, so the useful signal is qualitative: Dodge is adding people close to the product while it expands enterprise deployments.

That hiring pattern fits the technical problem. Context graphs and agents need software engineering, but production adoption also depends on integration work, evaluation, security review, and close observation of how support teams actually resolve exceptions.

Capital and Market Position

Accel and Google AI Futures Fund are the principal named backers in Dodge's $2.65M Seed. Schema Ventures, New Build Ventures, Antler, and angels from the SAP ecosystem also participated. Dodge has not disclosed its valuation, ownership terms, or a detailed allocation of the proceeds.

The investment arrives as enterprises are under pressure to modernize large application estates while preserving the custom processes that keep their businesses running. That pressure creates an opening between traditional application-maintenance services and general-purpose AI assistants.

Dodge is positioning itself in that gap. Its competition is not only another agent startup. It is also the consultant who knows the environment, the service provider that owns the support queue, and the internal team that does not want a new control layer introduced without evidence.

What to Watch

Dodge AI is worth watching because its product thesis matches a stubborn enterprise reality: the software is rarely the whole system. The real system includes years of decisions, exceptions, and human memory.

If Dodge can capture that context, keep it current, and let agents use it safely, application maintenance could move from repeated investigation toward cumulative learning. If it cannot, the control plane becomes one more abstraction that still needs an expert standing beside it.

The company has early capital, credible backers, a defined entry point, and company-reported deployment evidence. The harder phase begins now. Dodge has to show that institutional memory can become software without losing the judgment that made the memory valuable.