Too many companies dump millions into AI only to see 95% of those projects go nowhere. By 2026, this is shifting from a trend to a potential popping of the global bubble. We have the most powerful cognitive tools ever built, but they’re being used like blunt objects. The issue isn’t the machine. It’s the rigid, outdated structures we’re trying to force the machine into.
Trapped in Pilot Purgatory
“Pilot Purgatory” occurs when organizations get stuck in a loop of endless proof-of-concepts. They experiment forever but never scale. This is, unfortunately, a common problem that occurs when new technology such as Generative AI (GenAI) reaches a critical point on the hype scale. McKinsey found that 2/3rds of the companies experimenting with AI have not begun scaling across their organisation yet, and BGC research shows that only 5% of organisations are able to achieve any real AI value at scale. With investments hitting record highs, most GenAI projects are failing to show any real ROI. Why? Because we treat AI like a “bolt-on” accessory.
Our current workflows are not built for AI.
They’re linear because humans can only do one thing at a time, and that creates bottlenecks. When you add AI to these legacy processes, you are just making a bad system run faster, instead of resolving the underlying structural & architectural issues. You’re using a jet engine to help a horse carriage move faster when you should be building an airplane.
The Agentic Promise
The real win is moving from Generative AI to Agentic AI. Generative AI is an intern who can write a decent letter, but Agentic AI is an executive who can run the whole department. We’re moving from systems that “produce” to systems that “act.” This is the real revolution. Agentic AI is digital labour, where these systems plan, act and adapt on the fly. When the LLM stops being a chatbot and starts being an orchestrator, the whole geometry of work shifts.
In the old way, Task A had to finish before Task B could start, but with an agentic architecture, everything happens in parallel. Agents run dozens of sub-tasks at once and coordinate with each other. This isn’t just automation though; it could be a total reset of what productivity means.
Missing Consensus
The promise of AI appears to have some mixed results in the real-world. According to an IDC InfoBrief, organisations worldwide are realising an average of 3,7x ROI for GenAI investments. JP Morgan Chase have seen “a 10% to 20% productivity increase” from employees using LLMs for coding.
In the same breath, the McKinsey State of AI 2025 survey shows that only 39% of companies who experiment with AI are seeing EBIT impact at the enterprise level. Gartner projects that 30-40% of GenAI projects will be scrapped by 2027, without showing any real returns.
There is a lack of agreement on whether AI can consistently deliver on its promises, and if it can, we have yet to unlock the perfect implementation playbook that can scale across industries, use cases, and technology stacks. However, that presents an opportunity for businesses to discover the reward of massive competitive advantage if they can solve the AI-implementation problem.
The Leadership Mandate
-
Step 1:
If you're in charge, stop tinkering and start building. This means more than simply boosting your code-base & IT infrastructure. It’s about updating the culture and promoting controlled & managed experimentation with AI in all parts of your business. You need "AI literacy and fluency" across the board. You also have to integrate the data silos. A "Data Mesh" isn't a luxury anymore; it’s the baseline requirement to experiment and capture the lead. An agent is only as good as the data it can touch, and if your data is fragmented, your AI stays useless. -
Step 2:
Review your business processes and identify where the human bottlenecks exist. Without a solid understanding of where the opportunities lie, your AI experimentation is likely going to focus on a dead end before you even start. Good visibility on your opportunities allows you to minimise risk and claim the biggest returns. -
Step 3:
Scaled implementation should always be the goal. Even from day one, the ultimate goal of any AI experiment should be to gain returns that could be scaled. The value of AI is not in automating a single person’s tasks, but in creating a replicable system that can support the entire organisation. This is the difference between those who can and cannot achieve ROI at the enterprise level.
Stop Playing in the Sandbox
The era of the “AI pilot” is over. If you keep treating AI like a side tool, you’re going to get left behind. Success in 2025 requires a “built-in” philosophy. Redesign your roles for human-agent teams and build your systems for autonomy. We are shifting from the management of software to the management of a new kind of labour; one that can scale across the organisation & capture ROI in multiples that competitors currently cannot access. The sandbox was fun, but it’s time to formalise, commit, and start building the future of GenAI.
How Innovation Brewery can help:
Sources
- https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
- https://143485449.fs1.hubspotusercontent-eu1.net/hubfs/143485449/2024%20Business%20Opportunity%20of%20AI_Generative%20AI%20Delivering%20New%20Business%20Value%20and%20Increasing%20ROI.pdf
- https://br.nttdata.com/documents/2025/08/NTT-DATA-and-Cisco-InfoBrief.pdf
- https://www.digitalocean.com/community/conceptual-articles/build-autonomous-systems-agentic-ai
- https://blogs.microsoft.com/blog/2024/11/12/idcs-2024-ai-opportunity-study-top-five-ai-trends-to-watch/
- https://www.capgemini.com/wp-content/uploads/2025/07/Final-Web-Version-Report-AI-Agents.pdf
- https://www.bcg.com/press/24october2024-ai-adoption-in-2024-74-of-companies-struggle-to-achieve-and-scale-value
