How To Turn AI Agents Into A Full-Scale Workforce
Alexey Spas is the founder and CEO of Instinctools, a software engineering company focused on AI-powered digital solutions.
gettyWhenever I ask business executives, “How many AI agents are already running across your company?” the answer is usually some version of, “Honestly, I’ve lost count.” There are ones drafting sales proposals, handling customer support tickets, doing the coding and the list keeps growing. Individually, many of these agents are useful. Collectively, though, they are delivering nowhere near what organizations expected. By a 2025 MIT estimate, only 5% of solutions generate sustained productivity and P&L gains.
Most company leaders have seen the human version of this pattern in play: a team of experts, each excellent in their own lane, but without shared context and well-oiled processes; their strengths never translate into meaningful results. AI agents are prone to the same failure. Organizations create one impressive worker at a time, only to end up with a roster of digital employees stuck in Brownian motion.
Once the first AI agent proves useful, the instinct is to stamp out more of them. The growing number of agentic use cases may seem like progress, but when every agent is created in isolation, the only thing that scales is agent sprawl. And the same weaknesses get replicated with it:
Give an agent instructions but no business goal to measure against, and it’ll follow them to the letter, well past the point where a person would have paused to check.
Isolated decisions an agent makes based on its sliver of reality may turn out irrelevant at the big-picture level.
The more times work passes from one agent to another, the more context can get lost and the less efficient the agentic setup becomes.
Agentic systems respond to events instead of taking control over them.
As more agentic setups get diagnosed with these symptoms, it isn’t surprising that only one in four business leaders believes that AI agents will be able to act as autonomous workers in the short term.
The problem, though, isn’t the agents themselves. It’s what is missing around them.
I think of agentic systems like buildings: Each new agent adds another floor, and the higher you go, the more the structure depends on what sits underneath it. Here are six load-bearing capabilities I wouldn’t scale without:
This comes first, as agents need to work across the full software estate—cloud and on-prem, modern and legacy, from different vendors—without forcing the business into a single technology stack.
This situates agents within a unified understanding of the company’s processes, constraints and goals, so they can interpret data in terms of what it actually means for the business.
This coordinates the whole workflow, matching tasks to the right executors, whether a deterministic engine, an LLM or a human, and managing each handoff with a detailed audit trail.
Redesigned for a mixed workforce, these should standardize the work humans and agents share and measure the efficiency of their partnership.
Reusable components, such as agent templates, connectors, workflow patterns, tools and governance controls, keep implementation low-effort and low-risk.
Controls including smart routing, spend caps and live usage monitoring keep agentic workloads economically sustainable without compromising quality or security.
Get these must-win areas right, and you can add floor after floor without the whole thing wobbling.
Put these pieces together, and you get an agentic operating system: a shared orchestration and intelligence layer that allows individual agents to start behaving as one coordinated workforce.
That changes what agents can do. Instead of waiting for isolated prompts and acting on narrow slices of information, they can navigate complex enterprise environments, plan across multistep workflows, coordinate with other agents and people and act with a fuller understanding of business context. Feedback loops then allow the system to learn from outcomes and continuously improve how the work gets done.
Consider a claim from intake to payout. Running on the operating system, agents have a shared, business-aware view of the request, act on it and move through the process without losing context at the handoffs. Managers see where every claim stands and step in only to handle the edge cases.
The question that matters is no longer how many agents a company has, but whether there’s a unified operational layer underneath them. At that point, the investment case increasingly favors the connective layer that improves coordination, context and control across the agents already deployed. Once they run within an agentic operating system, the gains compound: Errors and incidents drop, every process becomes transparent, coordination between people and systems stays tight, and more human attention is reserved for the decisions that actually require it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
