When competence depreciates this fast, hiring for a fixed skill set is buying an asset that's already losing value. The durable advantage is the capacity to reskill, and most firms underfund it.
There's a quiet assumption baked into most hiring decisions: that the skill you're paying for will still be the skill you need in three years. For a long time that held. It no longer does. The half-life of a learned skill has fallen to roughly five years in technical fields and as little as two years in software development, which means a meaningful share of what a new hire knows on day one is obsolete before their second performance review.
That changes what a job offer actually is. You are not buying a capability. You are buying a depreciating asset, and depreciation that fast demands maintenance you have to fund deliberately. Treat training as a perk and you are letting your most expensive asset rust on purpose.
The contrarian read is uncomfortable but worth saying plainly. The constraint on AI return is rarely the model. It is the human who never learned to use it.
The numbers describe a maintenance crisis, not a hiring one
IDC projects that more than 90% of global enterprises will face critical skills shortages by 2026. Roughly 80% of the global workforce will need new skills by 2027. The World Economic Forum estimates that 59% of all workers, about 120 million people, require reskilling or upskilling by 2030, and that around 11% are unlikely to receive it at all. Demand for AI talent already outruns supply by about 3.2 to 1, with roughly 1.6 million open roles chasing some 518,000 qualified candidates.
Notice what these figures are not. They are not a story about a one-time talent gap that a good recruiting quarter closes. They describe a continuous erosion. The shortage refills itself as fast as you hire against it, because the skills keep expiring underneath the people who hold them. You cannot recruit your way out of a depreciation problem. You can only maintain your way through it.
A workforce is not a stock of skills you acquire once. It is a flow you have to keep replenishing, or it quietly runs dry.
Where the AI budget actually leaks
Here is the finding that should reorder a few budgets. Roughly one in fifty enterprise AI investments produces meaningful return. The reflex is to blame the technology, the integration, the data. Often the real cause is simpler and more embarrassing: the tool was bought, deployed, and then handed to people who were never trained to use it well.
Only about half of workers currently have access to adequate training. So picture the common sequence. A seven-figure platform arrives. Licenses are provisioned. And the upskilling line item, if it existed at all, was the first thing cut when the deal got expensive. The model performs exactly as advertised. The humans around it don't, because no one taught them to. The ROI was never going to materialize, and the post-mortem will name everything except the cause.
Figure 1 · How fast skills decay, by field
Field | Approx. skill half-life | Reskilling cadence required |
|---|---|---|
General workforce | ~7–10 years | Periodic; broad upskilling each cycle |
Technical roles | ~5 years | Structured annual refresh |
Software development | ~2 years | Continuous; built into the working week |
AI-specific skills | Under 2 years | Effectively rolling; faster than course cycles |
How to read it: As you move down the table, the shelf life of expertise shortens and the maintenance interval tightens. For AI roles the half-life is now shorter than most corporate training programs take to design, approve, and deliver.
Why training keeps losing the budget fight
Reskilling loses funding battles for a structural reason. Its cost is immediate and visible; its payoff is deferred and diffuse. A platform purchase produces a contract and a demo. A learning program produces a slow, hard-to-attribute lift in capability that shows up quarters later, by which point no one credits the training. So the line item that protects every other investment is the one most easily sacrificed to protect a quarter.
That logic is exactly backwards when skills depreciate this fast. The faster the half-life, the more the maintenance spend resembles a tax you pay to keep the rest of the workforce productive, not a discretionary benefit. Underfund it and the depreciation simply compounds in the background until the gap is wide enough to require panic hiring at a 3.2-to-1 disadvantage.
What this means for leaders
Budget reskilling as maintenance, not as morale. Move it out of the perks column and into the same category as keeping critical infrastructure patched. If a skill's half-life is two years, the spend that refreshes it is non-discretionary, and cutting it to make a quarter is borrowing capability you will have to repurchase at a premium.
Stop buying AI tools without an attached upskilling plan. When one in fifty AI investments returns and most failures trace to untrained users, the discipline is obvious: no platform purchase clears without a funded enablement plan and a named owner for adoption. The model is rarely the bottleneck. The person in front of it is.
Hire for learning velocity over present inventory. With WEF flagging 120 million workers who need reskilling and a chunk who won't get it, the candidate who adapts fastest will outrun the one who arrived with the perfect, perishable skill set. Screen for how quickly people acquire new competence, because that is the only attribute on the resume that doesn't expire.
The firms that pull ahead in this decade won't be the ones that hired the sharpest skills in 2026. They'll be the ones that built a reliable machine for replacing those skills before they expired, and funded it like they meant it. Everyone else will keep paying to rehire what they could have maintained.
A BusinessInfomatics original. Built on data from IDC, the World Economic Forum, and 2026 corporate-learning analyses, including Salesforce r
