Journal
Enterprise AI Agents: From Software Seats to Completed Workflows
Sep 29, 2026 · Donoghteh

Enterprise AI is moving beyond copilots toward systems that complete workflows. Here’s what founders, buyers, and investors should measure as value shifts from seats to outcomes.
The Enterprise AI Agent Test Is No Longer a Demo. It’s a Completed Workflow.
For years, enterprise software was sold by the seat: buy licenses, train users, and hope adoption translates into productivity. A new generation of AI companies is challenging that model by taking responsibility for work itself. The pitch is no longer simply “use this software.” It is “this workflow gets completed.”
That distinction is becoming more than a product story. It changes how companies buy, how founders build, and what investors should look for when evaluating enterprise AI.
From copilots to coordinated work
The first wave of generative AI in business often looked like an assistant layered onto existing tools: draft a response, summarize a document, or find an answer. Useful, but still dependent on a person to move each task forward.
Agent-based systems aim to connect the steps. A process that spans HR, finance, IT, and customer operations may require checking records, applying rules, updating systems, and escalating exceptions. The opportunity is to orchestrate that work across existing applications, while keeping people involved where judgment or approval matters.
This is not an argument that every workflow should be automated end to end. It is a more practical proposition: automate repeatable steps, make exceptions visible, and measure whether the whole process improves.
Capital is following the workflow
Recent funding offers a useful signal. TechCrunch reported on September 23, 2026, that enterprise AI startup Ema raised $77 million in a Series B led by Creaegis, bringing its total funding to $140 million. The company says it has more than 50 active enterprise deals and over one million enterprise users. It also reported that more than 90% of customers expanded beyond their first use case, and cited net dollar retention of about 180%.
Those figures are company-reported, and they should be read with the usual care. Still, the pattern is notable: the strongest evidence is not just initial experimentation, but customer expansion across workflows. That is where an AI product begins to look like an operating layer rather than a feature.
The same report describes a pricing model tied to task completion and business outcomes rather than seats or token consumption. That aligns the commercial conversation with the buyer’s underlying goal, while placing greater responsibility on the vendor to deliver reliably.
What founders need to prove
For founders, a compelling agent demo is only the beginning. Enterprise buyers will want evidence that a system can:
- Complete a defined workflow with reliable accuracy, not merely produce a plausible answer.
- Integrate with the systems of record that employees already use.
- Handle exceptions safely, with clear escalation and human oversight.
- Show measurable impact on cycle time, cost, service quality, or capacity.
- Earn trust through auditability, permissions, data controls, and predictable behavior.
A narrow starting point is often an advantage. One process with a clear owner, baseline, and success metric can establish credibility faster than a sweeping promise to automate an entire department.
What investors should measure
Seat growth and user counts remain useful, but they may not describe value creation fully when software performs work on a customer’s behalf. Investors should also examine workflow completion rates, exception frequency, time to production, customer expansion, gross margin after deployment support, and the durability of integrations and domain knowledge.
Retention deserves context too. A high net dollar retention rate can indicate expanding value, but investors should understand whether expansion reflects recurring production use, multi-year contract value, or a broader set of deployments. Bookings, annual recurring revenue, and realized revenue are not interchangeable.
The key diligence question is whether performance improves as deployments accumulate. If every new customer requires a large services team and bespoke engineering, growth may be hard to scale. If the product learns repeatable patterns while preserving customer-specific controls, expansion can become more efficient.
The new enterprise bargain
Outcome-oriented pricing sounds simple, but it shifts risk. Customers may pay for completed work rather than access, while vendors must account for the cost of models, integrations, monitoring, and human review. Contracts need to define what counts as completion, how exceptions are handled, and what happens when a workflow is blocked by inaccurate or unavailable source data.
That makes trust and operational discipline part of the product. Reliability is not a footnote to the demo. It is the mechanism that turns an agent into infrastructure.
The opportunity is real, but the bar is higher
AI agents do not make enterprise software disappear overnight. Existing systems still hold critical data, encode permissions, and support processes companies cannot casually replace. In many cases, the practical route is to work across those systems first, then let demonstrated performance determine what can be simplified or retired.
For founders, the opportunity is to own a valuable workflow and prove the result. For buyers, it is to start with a bounded process and measure the full operating impact. For investors, it is to distinguish durable deployment from impressive activity metrics.
The next chapter of enterprise AI will not be won by the agent that talks most convincingly. It will be won by the systems that complete important work, earn permission to do more, and make the economics clear to everyone involved.
Sources
- Jagmeet Singh, TechCrunch, “Ema raises $77M as AI starts eating into enterprise software and services,” September 23, 2026: https://techcrunch.com/2026/09/23/ema-raises-77m-as-ai-starts-eating-into-enterprise-software-and-services/
- Ivan Mehta, TechCrunch, “VC firm Bessemer now has another $5.75B to invest in (what else?) AI,” September 23, 2026: https://techcrunch.com/2026/09/23/vc-firm-bessemer-now-has-another-5-75b-to-invest-in-what-else-ai/