Last week, Anthropic disclosed that its AI assistant Claude now writes more than 80% of the code merged into the company's systems. Its engineers ship roughly eight times as much code per quarter as they did a few years ago. If you build training around a product like that, consider what happens to your content: the thing it describes changes faster than your review cycle can catch.
This acceleration isn't limited to AI companies. Every SaaS product with a continuous deployment pipeline, every regulated industry with rolling compliance updates, every organization adopting AI-powered tools is experiencing the same compression. The information in your training content is losing accuracy at a rate most L&D teams never measure.
That rate has a name.
The Clock Starts When You Hit Publish
The concept of a knowledge half-life comes from nuclear physics, borrowed by information scientists in the 1960s. In its original form, it describes the time it takes for half of the established knowledge in a field to be superseded or proven wrong. Fritz Machlup and others applied it to academic disciplines: engineering knowledge in the 1960s had a half-life of roughly ten years. Learn something as a freshman, and by your tenth year in practice, half of it would be outdated.
IBM's more recent research found that technical skills now have a half-life of about 2.5 years. In medicine, the numbers are more dramatic. Medical knowledge doubled every 50 years in 1950. By 2020, the doubling time had collapsed to 73 days, according to research published in Transactions of the American Clinical and Climatological Association.
But here's the part L&D teams miss: the knowledge half-life doesn't just apply to fields of study. It applies to every artifact you produce. Every onboarding module, every compliance course, every product walkthrough begins decaying the moment you publish it. Not because the content was wrong when you wrote it. Because the world it describes keeps moving.
Three Forces Compressing the Timeline
Not all training content decays at the same rate. A course on leadership principles has a half-life measured in years. A course on your company's expense reimbursement policy might hold for twelve months. A product training module for a SaaS tool on a two-week release cycle? Weeks. Sometimes days.
Right now, three forces are compressing half-lives across the board.
First, AI-accelerated development. When engineering teams use AI coding tools to ship eight times more code per quarter, product surfaces change at the same pace. Features get added, renamed, deprecated, and reorganized at a rate that makes quarterly content reviews feel like annual checkups for a patient in the ICU.
Second, regulatory velocity. Forty-eight states changed HR compliance rules in 2025 alone, according to ADP. Colorado's AI governance law (SB24-205) takes effect June 30, 2026, requiring impact assessments and risk management documentation for employers using AI in high-risk decisions. Each regulatory change can invalidate entire sections of compliance training overnight.
Third, production at scale without maintenance at scale. With 91% of companies increasing AI spending in L&D (Synthesia, 2026) and teams using AI for content drafting (60%), voice generation (63%), and video creation (52%), organizations are producing more training content faster than ever. More content means more surface area for decay. Production got automated. Maintenance didn't.
Annual Reviews Were Built for a Different Clock
Most L&D teams review content on an annual cycle. Some do it quarterly. A few ambitious teams manage monthly spot checks. But if your product training has a knowledge half-life of six weeks, an annual review means the vast majority of facts in that course will have shifted by the time someone looks at it again.
Software engineers wouldn't ship code for a year and then run tests. They run tests on every commit, every merge, every deployment. The feedback loop is continuous because they know that code breaks continuously.
Training content is downstream of that same code. When the code changes, the UI changes. When the UI changes, the screenshots in your product walkthrough are wrong. When the API changes, the workflow your course teaches no longer works. When the policy changes, the compliance scenario you built is misleading.
Yet most organizations treat content review as a calendar event rather than an event-driven process. The review happens when the quarter ends, not when the source material changes. By the time someone notices the drift, learners have already consumed inaccurate content, support tickets have already spiked, and in regulated environments, audit exposure has already accumulated.
From Calendar-Driven to Signal-Driven
Understanding knowledge half-life changes how you think about content maintenance. Instead of asking "when did we last review this?" you start asking "what has changed in the sources this content depends on?"
That reframing shifts the entire operating model. A calendar-based review treats all content equally: everything gets the same annual pass. A signal-based approach recognizes that a course tied to a rapidly evolving product needs monitoring at a different frequency than a course on workplace safety fundamentals. It connects training content to its upstream sources and watches for changes in those sources, not just the passage of time.
Software teams solved this decades ago with continuous integration. When a dependency changes, the build breaks, and someone fixes it before it ships. Continuity intelligence applies the same principle to knowledge: when a source document changes, downstream training content gets flagged, reviewed, and updated before learners encounter something outdated.
Measuring your content's knowledge half-life won't happen in a spreadsheet. It requires knowing what sources each piece of content depends on, when those sources last changed, and whether the content still reflects current reality. Only 10% of the $200 billion spent on corporate training annually delivers real results, according to research from the Association for Talent Development. A meaningful share of that waste comes from content that was accurate when it launched and inaccurate when it was consumed.
You can't stop knowledge from decaying. But you can stop pretending the decay isn't happening.
---
*Continuity Intelligence monitors your source documents and flags downstream training content the moment something changes, so your knowledge half-life becomes a metric you manage instead of a risk you ignore. Get your free drift report at https://continuityintelligence.com*
Enjoyed this article? Get more like it.
No spam. Unsubscribe anytime.
Your content is drifting right now. Let's prove it.
Paste a URL. Get a drift report. See exactly what's out of date — free.