Businesses are eager to adopt AI, no matter the price. Global blue-chip companies are making headlines for ‘token-maxxing’ through their yearly coding budgets in months. We shouldn’t be surprised to hear that Gartner expects AI coding costs to exceed the average developer’s salary by 2028.
This spending is changing who – or what – does the work. Organisations are moving towards a hybrid workforce – one where humans and AI agents work alongside one another. But, while this model creates exciting possibilities from productivity improvements to new levels of customer satisfaction, it is not a straight-forward cost-cutting solution.
Companies that introduced agentic AI to replace sections of the workforce are already running into issues. A 2025 Orgvue survey revealed that 55% of businesses regretted making employees redundant when bringing AI into the workforce. More recently, major manufacturers have been reversing job cuts and rehiring the employees they looked to replace.
In short, agentic AI shouldn’t simply be treated as a money saver. We also know that the technology will become more expensive in time, which creates new considerations for where and how agents are used. So, why haven’t “AI layoffs” been effective, and how can business leaders design an effective hybrid workforce that will last?
Why AI layoffs haven’t worked
For a couple of years now, we’ve heard claims from AI frontier companies about the huge workforce efficiencies that will be possible thanks to agentic AI. And maybe it’s not surprising that some businesses have drunk the Kool-Aid: whether it’s attempting to replace the HR department, or fully automate quality controls in manufacturing. It is already clear from several high-profile cases that this approach isn’t delivering, in turn forcing some companies to backtrack on these decisions. In some instances, organizations are having to rehire humans into customer-service roles, as customers felt the support they were receiving was too impersonal.
A major issue that’s coming to light is the importance of domain knowledge, usually gained through years of industry experience. AI can’t simply be dropped in without sufficient context and expected to replicate the kinds of results achieved by humans. The tacit knowledge that isn’t in databases, but in people’s brains, is extremely important for delivering consistent results long-term – but that’s exactly what’s missed when people are simply replaced.
It’s a similar scenario to when the Internet of Things was hyped: companies are showing up with this new technology, but forgetting to connect it to people – then realizing it doesn’t work. And for the companies pursuing AI layoffs, let’s not forget: AI models are non-deterministic. If you had a whole team of non-deterministic humans, you would fire them, because the output of that team would be too random.
Agentic AI needs to be built into the workforce, rather than instead of it, otherwise businesses will be left scrambling to rehire employees mistakenly replaced, at great cost.
How to design a hybrid workforce (not humans plus AI)
A major pitfall in agentic AI implementations is introducing AI like another technological add-on. But AI running in parallel with (or instead of) people is not going to work.
We need to consider how to build an effective hybrid workforce, where agents and humans collaborate on a long-term and sustainable basis. That means aligning humans and agents in several crucial ways – and treating them as a single workforce rather than two separate operational layers.
Firstly, we need to provide shared context: the same datasets and decision points. We need defined workflows, and the right application experiences.
If you’re just centralising raw data in a cloud lakehouse, you are not giving an agent sufficient context. AI cannot reason across a business by scanning disconnected, uncontextualised source records. To deliver actual value, data frameworks must map relationships between systems, assets, and human workflows.
Your organization’s AI outputs will only be as good as your data. Having a large volume of data is not an advantage if it is irrelevant or poor. It’s important to capture domain know-how through your ontologies, so that the answers provided by AI are actually statistically relevant.
Finally, it’s worth saying again that humans are essential to all this. Human to agent orchestration will be fundamental to making AI actually work in business critical workflows. So, a major part of the puzzle is working out where and how human intervention comes in.
Finding the right solution will require IT leaders to act as agentic architects rather than data gatekeepers. Autonomy cannot scale safely without boundaries and real-time oversight. Every digital interaction must leave an audit trail to keep the hybrid enterprise secure and compliant.
Ultimately, agentic AI alone won’t be the determiner of business success. After all, technology is increasingly democratised – a wide range of organisations can access frontier public models, as well as custom enterprise systems. This is not to say that everyone will use them well, though. The real secret sauce will be how successfully organisations can connect the technology to the people.
Focus on value, not cost-cutting
Leaders that see AI agents as a cheap replacement for human labor are doomed to be disappointed. It’s vital that from the beginning we think about agents as a virtual workforce with a budget. The balance between people and agents has to add up in terms of revenue targets, profit targets and margin targets – especially when you read all the latest stories about token-maxxing gone wrong.
Overall, this means thinking carefully about the value that your business provides – and where both humans and AI agents can deliver the greatest benefit. The hybrid workforce isn’t about cost efficiencies or using agents to automate the status quo, as organisations are learning to their cost.
Successful agentic enterprises will reimagine their business processes from the ground up – and find the best long-term balance of humans and agents for business success.
