The cost of more agents - The productivity mirage in AI
Founder hot takes in tech
Artificial intelligence was supposed to save us. By automating drudgery and accelerating analysis, it promised to free businesses from the grind of administration while creating time for more strategic work.
Instead, many firms are discovering that each new agent they add doesn’t lighten the load so much as piles it on.
Yet the agent economy is expanding fast. Sales teams install prospecting bots, finance departments deploy reconciliation tools and marketing divisions lean on digital copywriters and image generators. Each agent works well enough, but stitched together into an enterprise stack, they resemble a janky patchwork quilt. It’s expensive to maintain, prone to tearing and not exactly fit for purpose.
Rather than streamlining ops, the proliferation of agents risks entrenching the very inefficiencies they were designed to fix.
The illusion is powerful. A chatbot that answers queries or a model that drafts memos creates the impression of time saved. Yet with every task delegated, the chance of error multiplies. A mis-summarised report cascades into a poor decision, a hallucinated fact becomes the basis for a misguided investment. Bias compounds silently.
This is what might be called the ‘productivity mirage’ i.e. the gap between AI’s promise and its present performance. Despite billions invested, productivity growth across advanced economies has barely flickered. The OECD reports no discernible uptick. McKinsey estimates adoption plateaued at roughly half of firms. Accenture finds 40% of pilots never scale beyond experiment. For many executives, however, the only obvious outcome of AI to date is better subject lines on marketing emails.
The mirage is particularly striking in small and medium-sized enterprises (SMEs), which form the bulk of most economies. In Decidr’s National AI Readiness Index 2025, nine in ten Australian SMEs already use AI tools, but only 39% feel confident in implementing them effectively. Usage is largely confined to generative systems such as ChatGPT or Copilot, which are helpful for tasks but rarely decisive in strategy. The gulf between belief and impact is wide. 83% of leaders expect AI to transform their businesses within a year, yet fewer than half have made it a genuine priority.
This disconnect explains why AI so often feels like sand through the fingers. Businesses sense its potential yet struggle to capture it. The problem is not a lack of ambition but of architecture. Adding more agents is like adding more taps to an empty pipeline. Plenty of flow, but nothing connected.
As I suggested at the recent talk I gave on AI sovereignty, AI doesn’t fail because it’s weak. It fails because it’s lonely. Agents working in isolation become expensive novelties. Integration, not proliferation, is the precondition for productivity. Without a system that governs and connects them, firms risk death by a hundred (thousand?) bots.
The stakes are high. The average enterprise already runs close to 200 apps, with employees losing much of their time to “swivel-chair” work. Shifting data between systems that refuse to talk. Introducing a dozen more agents without a unifying layer merely deepens the maze and the risk is that AI becomes another silo, bolted on rather than built in.
So what’s the fix?
What is needed is an agentic layer. A system that does not simply add agents but orchestrates them. Instead of dozens of point solutions, a single horizontal operating system can connect data, govern decisions and run workflows end-to-end. This is the difference between AI as a tool and AI as infrastructure.
And, rather conveniently, that is the logic behind DecidrOS. Designed as the first agentic AIoperating system, it acts as a business brain rather than another app in the stack. CRMs, spreadsheets, finance systems, customer platforms are all stitched into one schema. Instead of a chatbot that drafts a follow-up, DecidrOS sends it. Instead of a tool that flags a reconciliation, it finalises it. Instead of a report generator, it executes the plan.
The dividend is not novelty but efficiency. The outcome is small businesses reclaiming 20 hours a week, enterprise teams cutting “swivel-chair” work nearly in half and leadership confidence finally embracing the technology and soaring. Productivity gains, in other words, banked as bottom-line growth.
Until businesses embrace such integration, adding agents will keep adding dollars, but with an agentic layer, however, the Productivity Mirage begins to fade and AI’s promise comes into view.
