Where the Agent Actually Starts
Accenture Ventures announced a strategic investment in Within on September 23. The financial terms were not disclosed. What is disclosed is the bet: that the hardest part of deploying AI agents inside a large company is not the model, the framework, or the vendor platform. It is knowing how work actually gets done.
Within’s platform captures how employees work across applications and tracks offline interactions — the undocumented handoffs, exceptions, and workarounds — then compiles that into what the company calls a ‘Work Brain’: a continuously updated context layer that AI agents draw on to execute tasks across finance, HR, sales, IT, supply chain, and other functions. The platform is built on private, sovereign architecture, aimed at regulated industries where data residency matters.
Andrew Antos, Within’s CEO and co-founder, frames the pitch this way: “AI transformation works best when there’s a system of record of work to provide visibility on what to automate and the ability to verify that agents are doing their work the right way.”
The 82-23 Problem
That pitch arrives at a moment when the enterprise AI market is running into a measurement wall. Accenture’s Pulse of Change survey — over 3,000 C-suite leaders, and worth noting that Accenture is the party commissioning and publishing its own research here — found that 82% of organizations are increasing their AI investment. But only 23% report achieving widespread, sustained business value from it. That number was 32% earlier in 2026. It is going in the wrong direction.
The gap is not primarily a technology problem. It is a visibility problem. Jason Dess, Accenture’s Industry and Process Reinvention Lead, puts it plainly: “One of the most persistent challenges in AI transformation is that organizations often don’t have a clear picture of the starting point — the actual processes, the exceptions, how work actually flows through their teams. Without that understanding, it’s difficult to scale AI beyond isolated use cases.”
This is the gap Within is designed to fill. David Benjamin, EVP and Chief Commercial Officer at Blackbaud — an existing Within customer — describes what the platform surfaced for his team: “What Within surfaced was the real process, not the one on the diagram — the hidden workarounds, the manual steps, the informal patches our people had built to get the job done. You can’t reinvent what you can’t see.”
Three Ways to Deploy an Agent
The Accenture-Within deal is not just an investment. It is the formalization of a channel. Three distinct deployment models for enterprise AI agents are now in play.
The first is vendor-packaged: agents embedded inside existing SaaS platforms. Salesforce AIforce at Dreamforce positioned the agent as the enterprise UI itself. ServiceNow has moved similarly. The agent lives inside the software you already pay for.
The second is DIY: internal teams building on open-source frameworks like LangGraph or CrewAI. More control, more maintenance burden, and the measurement and governance problems land on the enterprise directly.
The third — the one Accenture is formalizing — is SI-deployed. Buy the platform. Let your systems integrator deploy it, configure it, manage it, and own the ongoing relationship. The economics here are different from pure SaaS: implementation fees, managed services, and the kind of deep enterprise lock-in that comes from a consultant capturing your actual workflows inside their system.
Who Controls the Process Layer
This is where the stakes get clearer. In the vendor-packaged model, the software company controls the agent and the data. In the DIY model, the enterprise controls everything but carries all the complexity. In the SI-deployed model, the consultant controls the process-capture layer — the ‘Work Brain’ that determines what the agent knows and how it operates.
That is a different kind of lock-in than a software subscription. The agent’s intelligence becomes tied to the SI’s specific mapping of your workflows. Switching means rebuilding not just software configuration but an entire operational picture of how your company works.
Accenture Ventures is not doing this for one company. It is a $250 million fund with over 70 active venture investments, making 10 to 15 deals a year. AI agents are a core thematic focus — recent bets include Netomi (agentic CX), AlphaSense (agentic market intelligence), Replit (AI coding agents), Netail (retail AI agents), and Lyzr (agentic AI infrastructure). The Within investment is the latest in a pattern.
What Remains Unproven
The SI channel sits in contrast to the “replace SaaS entirely” pitch that Ema made with its $77 million Series B, where outcome-based pricing targets the per-seat model directly. It also sidesteps the measurement problem — five competing metrics, no standard — by placing the proof burden on the consultant rather than the vendor. And it exists in a world where agent identity is still fragmenting across vendor-specific silos, as Forkast’s recent coverage of the identity stack consolidation examined.
What the evidence does not yet show is whether the SI-deployed model actually closes the 82-23 gap or simply redistributes the cost and control. All company-sourced claims about Within’s platform — the Work Brain capabilities, the Fortune 500 customer base, the breadth of functional coverage — come from the Accenture announcement and have not been independently verified. The Accenture survey data that frames the problem comes from Accenture’s own research, and the investment thesis naturally serves Accenture’s positioning as the SI that deploys this technology.
For the average employee inside one of these companies, the question is simpler than the deployment model. Whether the consultant or the vendor controls the agent, the test is whether these systems actually reduce the friction of daily work — or add another layer of managed complexity on top of the workarounds people already built to get things done.