Operating knowledge stopped being a search problem on October 28. In one day, a startup raised $56 million to turn it into autonomous action, two platform companies wired it into an assistant that executes workflows, and a product data company put agents to work on thousands of records at once. The context layer is no longer only answering questions. It is starting to act, and that changes what a wrong record costs.

Operating knowledge leaves the head and enters the engine

Fragment launched on October 28 with $56 million in total funding, including a Series A led by 8VC and a seed round co-led by General Catalyst and A*, according to its launch announcement. The company describes itself as building software that turns operating knowledge into autonomous action, starting with direct procurement at large physical goods companies.

At the center sits its Autonomous Context Engine, or ACE, which connects information across separate systems and captures the company specific policies, relationships and institutional knowledge that experienced employees rely on. Fragment says its agents use that understanding to investigate purchasing and payment exceptions, determine the appropriate response and carry out authorized resolutions end to end, rather than flagging issues or recommending next steps. The company says it is already working with Graybar and CNX.

The problem it names is familiar. Order changes, invoice discrepancies, shortages and tariff changes force employees to piece together information across systems and teams, often relying on institutional knowledge and unwritten business rules. That hidden operating labor is exactly the material a company record holds, when a company has written it down at all.

CEO Pranav Mulgund put the ambition plainly in the announcement: procurement teams have spent too much time chasing discrepancies across systems, and AI now makes it possible to build software that understands the business and carries that work through.

The suites made the same move on the same day

ServiceNow and Google Cloud announced an expanded partnership on October 28 that connects what the release calls the industry specific operational context of ServiceNow Action Fabric to Gemini Enterprise, in the joint announcement. Gemini Enterprise will be able to reason and take complex, multi step actions using ServiceNow workflows, connected through Model Context Protocol servers and Agent2Agent connectivity.

The retail, customer relationship management and IT integrations are generally available to mutual customers now, the release says, while the telecommunications integration is expected in early 2027. Sellers can ask for account context and then create follow ups in the same interface. Retail teams can report equipment problems, and field service can dispatch a technician.

ServiceNow executive John Aisien stated the shift in one line: conversational AI is only as useful as the action it drives. That sentence describes the whole day. Context is valuable now because an agent can spend it.

Acting raises the price of a stale record

Akeneo made the control version of the same argument on October 28, in its Autumn Release 2026. Its agents can enrich thousands of products, onboard supplier data and publish to every channel at once. The release stresses that this reach is why control matters: its agent Ziggy works within the same role based permissions as the human team, records every action as a traceable change, and proposes changes that a person approves before anything is applied.

CEO Romain Fouache said agents that act on product data driving pricing, compliance and brand have to act under control, inside a harness with permissions, traceability and human approvals.

The contrast between the two launches is the design question of the next year. Fragment leads with authorized resolutions carried out end to end. Akeneo leads with a proposal a person approves before it applies. Both depend on the same thing: the record underneath has to be right, current and checkable.

A wrong answer costs a correction. A wrong action costs more. An answer based on a superseded policy can be caught by a reader. A resolution based on that same policy changes a payment, a purchase order or a dispatch before anyone reads anything. That is why the record questions this site has tracked in evolving workplace records and in shared AI context now carry operational weight, not only search quality.

What an acting agent needs from the record

An agent that acts needs more from operating knowledge than relevance. It needs to know where a rule came from, when it was set, which version it replaced, and who approved it. It needs a correction to reach actions already queued, not only answers given later. And it needs a clear line between what it may resolve on its own and what it must propose, which is the same review gate described in how Jilu works and in the context engineering pillar.

October 28 suggests the market has settled the first question. Operating knowledge will be captured in engines and spent by agents, by startups and by suites at the same time. The open question is the second one: whether the record those agents act from is a reviewed record a person can open, date and correct, or a summary no one can inspect until an action goes wrong. Companies that get the record right will let agents act with confidence. Companies that skip it will automate their oldest errors at a speed they have never managed before.