Summary

• Conduct has raised a $60 million Series A from investors including Index Ventures, Iconiq, SAP, Creandum, Lucid Capital, and Bloom.
• The startup is led by former Palantir engineers and is targeting the operational layer between AI agents and enterprise systems.
• The story sits in the gap between agentic AI demos and the practical work of integrating agents with ERP, workflows, permissions, and legacy software.

London based Conduct has raised $60 million to build an agentic operating system for large enterprises, adding another heavily backed UK company to the race to make AI agents work inside complex organisations rather than controlled demos.

The Series A was backed by Index Ventures, Iconiq, and SAP, with Creandum, Lucid Capital, and Bloom also participating. Conduct was founded by three former Palantir engineers and is targeting the layer where agents interact with enterprise applications, data, workflows, and business logic.

A reliable official company website could not be verified during checks, so the company name is not linked in this draft.

Enterprise AI attention is shifting from standalone copilots to agents that can perform tasks across systems. In theory, those agents can update records, prepare reports, check compliance, manage workflows, coordinate actions, or support customer operations. In practice, they meet the same constraints that have shaped enterprise software for decades: messy data, customised systems, permissions, audit trails, process exceptions, and legacy integration.

Conduct appears to be targeting that operational layer. Large organisations rarely run on clean software estates. They use ERP systems, CRM platforms, spreadsheets, workflow tools, data warehouses, procurement systems, sector specific applications, and internal software customised over many years. An AI agent that performs well in a controlled environment can fail when it meets conflicting data formats, approval chains, incomplete permissions, or business rules that exist because previous processes broke and were patched around.

SAP’s involvement is notable because enterprise AI will depend heavily on systems of record. ERP platforms hold financial, supply chain, human resources, procurement, and operational data. Agents acting inside those environments need to respect authorisation, segregation of duties, compliance, auditability, and rollback.

The commercial opportunity is large, although implementation will be expensive and uneven. Many AI agent companies present automation as a model capability problem. In enterprise environments, models are only one part of the stack. Agents need context, permissions, tool access, identity, monitoring, governance, and integration with processes that may differ across departments, geographies, and acquired businesses.

Conduct’s Palantir connection helps explain investor interest. Palantir built much of its business around integrating complex data and operational systems for governments and large enterprises. That background is relevant to agentic AI because the difficult work is often not generating an answer, but mapping an organisation well enough for software to act usefully and safely.

The UK angle is also strong. London combines technical talent, financial services demand, public sector complexity, and access to European buyers. Conduct’s large Series A suggests investors still see room for European companies to build infrastructure around enterprise AI, even as frontier model development remains concentrated in the US.

The risk is that “agentic operating system” becomes a broad category label without enough clarity for buyers. Enterprises will want to know what Conduct controls, which systems it integrates with, how it handles errors, how it audits agent actions, and whether it reduces implementation cost compared with existing automation, integration platform, and workflow tools.

The round is best read as a bet on operational AI. Companies are not short of models or demonstrations. They are short of ways to connect AI systems to the messy, governed, and regulated environments where work actually happens. Conduct’s challenge is to prove that an agentic operating layer can make that connection without becoming another expensive integration project.