Outcomes over Assistance

Why we’re less excited about co-pilots and far more interested in agents at Pruven Capital

Generative AI has delivered a wave of “co-pilot” solutions and applications—software that acts as an assistant by providing suggestions, partial work products or components, and other forms of support to human (and largely knowledge) workers. At Pruven Capital, however, we are skeptical about the long-term viability of many businesses based solely on a co-pilot model. We are far more interested in fully agentic AI solutions that can fundamentally transform rather than augment how businesses and functions operate.

Friendly reminder: Humans (and organizations) are intrinsically lazy

Co-pilots promise to offer knowledge workers quicker document reviews, generation of analyses, faster code generation, smoother design iterations and other productivity enhancements. Yet despite several studies claiming that co-pilots deliver improvements of 25-30%, few of our financial services LPs or other enterprises we’re familiar with have seen these productivity gains manifest in the top or bottom line.?

Our hypothesis: most of the time saved accrues to employee time on TikTok, Insta or e-Tail. As a quick thought experiment (and be honest): when a meeting on your calendar is canceled with little notice, how much of that liberated time goes to other productive tasks vs. taking a breather?

For co-pilots to gain at-scale traction, they must demonstrate substantial improvements in existing workflows done through existing operating models. The bar for elevating the resulting work products and services is often surprisingly high, particularly within large organizations that are already reasonably optimized for efficiency and consistency. One of our LPs trialed a leading financial services co-pilot, and while the employees universally agreed that the solution was good, few felt it was worth the monthly per-user SaaS fee charged. The work was already being done effectively; a “bit better” wasn’t worth adding incremental cost.

Co-Pilots Are the Next “Search Bar”

We also anticipate that in the next three to five years, co-pilot features will be standard across virtually all applications, much like the ubiquitous “search bar” found in software today. While many of these AI-powered assistants will undoubtably be helpful, we worry again that the incremental support they provide won’t clear the hurdle to drive monetization. Instead, we see a future where co-pilots become a basic feature that is generally commoditized.?

Why Agentic Solutions Are the Future

We believe the real potential of generative AI lies not in incremental enhancements to human-led work, but using agentic solutions to transform how work is done and delivered. Instead of acting as a “co-pilot” that requires continuous human guidance, agentic AI solutions act as independent and autonomous drivers, capable of managing tasks end-to-end with minimal intervention—with the work done faster, more consistently, more effectively and at radically lower cost. Realizing the value from these solutions requires fundamentally reimagining not only the workflows, but the surrounding operating model and organizational structure.

This transition is hardly radical or new for enterprises; consider the rise of business process outsourcing (BPO) over the last 50 years. When companies outsource tasks, they contract with a BPO provider to fully take over the work, delivering results according to specific quality, responsiveness, uptime and other requirements. The BPO provider doesn’t “ride shotgun”; it delivers completed tasks for a contracted cost with clear expectations and accountability. Agentic AI solutions should work in much the same way. Just as a BPO provider creates efficiency by allowing companies to “hand off” certain functions entirely, agentic solutions will enable businesses to hand over specific tasks or processes to AI-based solutions. In turn, enterprises can restructure and reallocate resources accordingly.

The Business Model of Agentic Solutions

Therefore the success of Agentic AI companies will hinge not only on the capability of their technology but also on the business models they employ. Winning agentic companies will need to go beyond selling software licenses or usage-based subscriptions. Instead, they should offer contracts that align their revenue with the outcomes they deliver, providing tangible value to their clients, including:

• Service-Level Agreements (SLAs): Just as BPO providers adhere to SLAs, agentic AI solutions should offer clear service guarantees around the work they deliver. This means defining specific benchmarks for performance, responsiveness, uptime, quality and other metrics, ensuring that clients can trust the AI to meet essential operational standards (and recourse if they don’t)
• Impact-Based Revenue Models: The most successful agentic companies will likely tie a portion of their revenue to the outcomes they help their clients achieve. For example, an agentic solution deployed in a customer service role might structure its fees based on customer satisfaction ratings or response times. This kind of impact-based model creates accountability and aligns the AI provider’s incentives with those of their clients.
• Organizational Blueprint and Transition Roadmap: Transitioning to agentic solutions isn’t just a technological shift; it’s an operating model transformation. Successful providers will not only offer a solution; they’ll also provide a blueprint for how an organization’s operating model will evolve, including anticipated changes to employee roles and headcount. This roadmap should provide clients with a clear picture of how the new technology will be integrated, milestones and proof-points on the journey, and a clear picture of what the end-state of their organization will look like.

Caveats

We recognize and appreciate that the lines between Co-Pilots and Agentic solutions are blurry at best, and are more fairly represented as a spectrum of capabilities. We also know that Co-Pilots are often a necessary stepping-stone toward developing a fully realized Agentic model. We also believe that some highly sensitive and / or intrinsically human functions may never be suitable for Agentic solutions. All that said, we would encourage Founders to drive toward Agentic capabilities as far as they can—recognizing that in most instances, the ideal end-state for their customers will keep moving in that direction.

Outcomes over Assistance

Both Founders and Enterprises that are ready to embrace delivering outcomes over assistance—and all that entails not only in the underlying technology but also the offering and business model—are the ones that will realize the full promise of GenAI. Please reach out to us if you feel the same way.