AI Skill Half-Life Engine.
How fast each AI skill's premium is decaying. Three decay shapes: exponential, logistic, plateau. Half-life. Inflection date. Forecast which skills will hold their premium and which won't.
What it is, and why it exists.
The Skill Half-Life Engine answers a question every engineer and learning team is asking out loud right now: if I invest two hundred hours learning this skill, how long will the premium last? The Skill Premiums engine tells you what a skill is worth today. The Skill Half-Life engine tells you how long that value persists.
AI skills decay faster than skills in any other domain we track. Prompt engineering commanded a $30-50K premium in 2023; by mid-2025 the basic-tier premium had compressed to under $20K as supply caught up. RAG ops carried a $90K premium in 2024; by Q4 2025 it had fallen to $42K because production RAG was no longer the moat it used to be. CUDA kernel optimization, by contrast, has held its premium for years because the underlying skill is hard and supply stays scarce.
The engine takes the premium history for every skill in the catalog, fits one of three decay curves (exponential, logistic, plateau), and returns four fields: half_life (in years), curve_shape (one of the three), inflection_date (when decay is fastest, for logistic curves), and methodology_version. Every field comes with the standard Bloomberg-grade transparency: sample size, confidence interval, source breakdown.
This is not a forecasting engine in the sense that we predict the future market. It is a fitting engine: it tells you what the rate of change looks like and projects it forward under the assumption that the rate continues. That assumption breaks during structural shocks (a new model class, a hiring freeze, a regulatory shift). We flag those moments with structural_shift_detected: true on the response.
Three decay shapes.
Forcing a single curve shape onto every skill produces bad fits. Some skills decay exponentially. Some hit an inflection and bend. Some plateau. The engine fits three candidate curves to each skill's history and picks the one with the best out-of-sample log-likelihood on a rolling 90-day evaluation window.
Constant percentage decay per unit time. Once supply starts catching up to demand, the premium falls on a steady curve.
S-curve with an inflection point. Premium holds while supply is scarce, then bends sharply once supply hits the inflection.
Slow long-tail decay. The skill is hard to learn or hard to acquire; supply stays scarce even as demand normalizes.
Every response includes the curve shape, so customers can build their own scenarios. A learning team can plot the curve, mark today on the x-axis, and answer the question by the time we finish training our engineers on this skill, how much premium will be left. Without the curve shape, the half-life alone could mislead: a logistic curve with an inflection in six months looks very different from an exponential curve with the same nominal half-life.
Sample half-lives.
Snapshot from the May 2026 fit. Half-life is years until the current premium falls to 50% of present value under the fitted curve.
Methodology version 2026.2-half-life-preview. Engine in preview status: half-life estimates for skills with under 18 months of history carry wider confidence intervals.
How buyers use it.
The most common use case. Engineers compute the expected value of a skill investment by multiplying the current premium by the time-to-decay-to-half. A skill with a 5-year half-life and a $60K premium offers ~$200K in lifetime premium under typical career durations. A skill with a 1-year half-life and a $60K premium offers less than half that. The half-life lets engineers stop chasing every shiny framework and concentrate on durable skills.
Internal learning teams use the curve to time their training programs. Teaching a skill whose inflection is six months away is a poor investment for an 18-month upskilling plan; teaching one with a 4-year plateau is a great one. The engine's API exposes the projected premium 6, 12, 18, 24, and 36 months out for any skill, so curriculum planners can pull a forward-looking comp impact for every module.
Pay pressure does not just come from base salary inflation. It comes from the premiums on the specific skills your engineers already hold. If your team is heavy in a skill whose half-life is under two years, you can expect that team's comp expectations to compress as the premium decays. The engine surfaces this 18-36 months in advance.
Through the MCP analyze_skills tool with include_half_life: true, agents get both the current premium and the curve. The Decision-Ready Response carries an explicit limitations array listing skills whose half-life carries low confidence, so the agent never recommends a bet without flagging the risk.
Methodology.
For each skill, we have a time series of quarterly premium estimates from the Skill Premiums engine. The Half-Life engine fits three candidate curves to that series and selects the best fit by out-of-sample log-likelihood on a rolling 90-day evaluation window.
The curve choice is part of the response. The customer never has to know about the fitting machinery: they just get a half-life, a shape label, and an inflection date where applicable. But the methodology is fully documented at the methodology page so anyone can audit the choice.
Structural shifts: a new model class entering production, a sudden hiring freeze, a regulatory change that creates demand: are detected when the most recent two quarters of premium data fall outside the 95% predictive interval of the fitted curve. When that happens the response includes structural_shift_detected: true and a follow-up suggestion to refit. We do not silently smooth over inflection points; we surface them.
API integration.
Two integration paths. Both return the same envelope.
Full reference at /orbyt-intelligence/api-docs. The MCP server documentation is at /orbyt-intelligence/mcp.
Honest limitations.
- Preview status. Many AI skills have fewer than 24 months of premium history. Fits on those skills carry wider CIs. We flag low-confidence fits with
disagreement_flag: trueand exclude them from the headline table on this page. - Curve assumption. The engine assumes premium will continue along the fitted curve. New model classes (GPT-5, Claude 5, whatever comes next) can reset the curve completely. We surface structural shifts when detected, but cannot predict them in advance.
- U.S. only. Premium history is U.S.-only as of May 2026. International coverage rolls out with per-region methodology versions in 2026.
- Aggregate, not personal. Half-life is a market-level estimate. Your individual premium may decay faster or slower depending on your specific specialization. The engine returns the market median; not a guarantee for any one engineer.
Pricing and access.
Half-life queries require the Pro tier and the skills:read scope. While in preview, half-life data is offered at the same rate as Skill Premiums. The engine moves to v1 stable in Q4 2026 at the latest; pricing structure will not change at that transition.
Full pricing detail at /orbyt-intelligence/pricing.
See also.
Methodology version 2026.2-half-life-preview. Last updated May 2026. Engine in preview status: moves to v1 stable in Q4 2026 once the average skill has 24+ months of history. Structural shifts surfaced with structural_shift_detected.