AI can make the creation of work dramatically faster. Unless L&D helps their organisation to change how work moves, gets judged and turns into value, the result will be pressure, not progress: stress on teams that are already at capacity, using process that weren’t designed for AI, says Erica Farmer.
I keep hearing a version of the same story; someone uses AI to produce a first draft, analysis or set of options in a fraction of the usual time. Their manager is delighted. Then, a few weeks later, the team feels busier than before. There are more documents to review, more ideas to discuss, more recommendations waiting for a decision and more work that looks ‘good enough’ but has not made anything better.
That’s what I call the AI fire hydrant. The flow of work has increased, but it is still being forced through the same old pipework: the same meetings, approval layers, decision queues, quality checks and overstretched managers. Somewhere in the system there is usually a tiny O-ring, often a conscientious manager or quality team, expected to review, challenge and sign off twice the volume that’s arriving at twice the speed.
Can the organisation absorb the new capacity?
This is not a marginal issue for L&D. It is rapidly becoming the real work of AI enablement. The question is no longer simply, ‘Can people use the tool?’ It is, ‘Can the organisation absorb the new capacity in a way that improves outcomes rather than creating more widgets?’
Why individual gains do not automatically become organisational gains
The evidence is beginning to catch up with what many of us can already see at work. Demirer, Musolff and Yang’s 2026 research, aptly titled Writing Code vs. Shipping Code, makes the point beautifully: speeding up the production step does not guarantee that the finished product reaches the customer faster. The constraints often sit elsewhere in the system.
The Organisation for Economic Co-operation and Development (OECD)’s 2025 review of research on generative AI, The Effects of Generative AI on Productivity, Innovation and Entrepreneurship makes a similarly important point: results vary by task and user experience and human–AI collaboration is central to turning potential into real value.
And Microsoft’s 2026 Work Trend Index describes a group of skilled employees being ‘blocked’ because organisational readiness has not caught up.
That is the danger: fast people inside slow organisations.
So, please stop your people treating self-reported time saved as the final KPI. It is a useful signal, but it is not a return on investment. If an employee saves 30 minutes and that time disappears into fragmented requests, unnecessary checking or a queue for approval, the organisation has not created 30 minutes of value. It has merely moved the pressure downstream to someone else.
The better question is: what happens to the released time, energy and capacity? Does it improve customer response, reduce rework, enable better judgement, create space for learning, time back for personal items, or allow a team to stop doing low-value work? That is the AI Dividend™ worth designing for.
Here are four practical jobs for L&D which we can do right now:
- Map one real workflow, not a theoretical future state
Pick a high-volume activity; a client proposal, policy update, learning design brief, recruitment shortlist or management report. Follow it from request to outcome.
Where does it wait? Who touches it? Which reviews genuinely reduce risk, and which exist because nobody has revisited them? Do this with the people who perform the work and the people who receive it.
L&D can bring the facilitation discipline that turns vague complaints about ‘being busy’ into a visible workflow problem. - Create team-level norms for good AI-enabled work
Policy matters, especially data, risk and accountability. But policy does not answer the everyday questions that make managers the bottleneck: what needs human review; what evidence should accompany an AI-assisted recommendation; when can someone act without permission; how do we show the source, assumptions and limitations; and what quality threshold is actually required?
Make these norms role specific. A blanket expectation that every AI output must be checked in the same way is neither safe nor scalable. - Develop managers as capacity designers, not just tool adopters
This is the bit we are often avoiding. Managers need more than prompt training. They need permission and practical support to remove redundant meetings, reset approvals, clarify decision rights and say, ‘we do not need a report on that anymore’.
Give them short, facilitated redesign sessions built around their own work. The outcome should be a few changed habits, owners and measures, not another slide deck. - Make quality discernment a core capability
Anthropic’s Economic Index shows that AI use is evolving, with people increasingly delegating tasks to AI. That makes human judgement more, not less, important.
L&D should teach people to challenge outputs, spot omissions and bias, test evidence, use domain expertise and decide whether the work is worth doing at all.
AI drafts; humans craft. The premium shifts from producing a first version to setting the right question, applying judgement and being accountable for the decision.
Measure value where it lands
L&D should help leaders measure the workflow, not just the learning. Before an intervention, agree two or three outcome measures that matter, such as:
- Time from request to decision
- Number of approval hand-offs
- Rework rates
- Customer response time
- Error rates or hours returned to higher-value activity
Revisit them after 30 and 90 days. Ask the team what they have stopped, simplified or improved as a result.
This is also where L&D plays a much more strategic role. We are not simply the team that helps people get confident with new technology. We can be the function that notices when capability is colliding with an outdated operating model, brings the right people together and helps turn individual experimentation into healthier, more effective work.
We have already unleashed the AI fire hydrant. The organisations that benefit will not be the ones that produce the most, they will be the ones brave enough to widen the pipework, remove the unnecessary valves and decide deliberately where the new human capacity should go.
Erica Farmer is AI and Future of Work Speaker, Trainer and Co-founder at Quantum Rise and author of AI For People Professionals
Erica is also keynote speaker at the TJ Conference 2026
This article was created by Erica Farmer in collaboration with GPT‑5.6 in ChatGPT Work, which supported the development and refinement of her original ideas
If this article piqued your interest, you’ll probably find value in our conference!
Find out about keynote speaker Erica Farmer, her opening session on the
AI Fire Hydrant and what it means for productivity, performance and L&D
Don’t miss 50% off your ticket for the first 30 only!
