AI News Digest, April 17: The AI Training Gap No One Is Solving

Companies have been running AI training programs for two years. The results are in, and they’re not good. A new Docebo report out this week finds that 85% of employees cannot apply their AI training to their actual daily work. Not because of resistance. Because the training itself is broken. The AI readiness training gap is widening even as the tools your teams are supposed to use get more capable by the week. OpenAI just upgraded its Agents SDK for enterprise safety. Salesforce launched a unified AI marketplace with a $50M builder incentive. And Vietnam became the first ASEAN nation with a binding AI law. Here’s what moved this week.

The AI Training Gap That’s Costing Your Company

Docebo, in a survey of 2,000 enterprise employees conducted by research firm Centiment, found that 85% cannot apply the AI training they’ve received to their actual job. The study covered employees and L&D leaders across the US, UK, Canada, France, Germany, and Italy — and found AI literacy ranks as the top L&D priority for both groups over the next 12 to 18 months. (Source: HR Dive | Full report: Nasdaq)

Let that land. Four out of five employees sit through AI training and walk out the door unable to use what they learned. The money and time are going in. The behavior change is not coming out.

Why? Docebo’s report identifies three structural failures behind the AI readiness training gap. Nearly 60% of employees say their organization’s learning programs aren’t designed with people like them in mind. One in five hasn’t received any AI training at all. And 56% are so buried in pre-AI manual tasks that they don’t have time to apply new tools even when training is available.

This is a change management problem dressed up as a training problem. Most companies run generic, one-size-fits-all AI literacy programs. Learn what a large language model is. Try a few prompts. Go back to your spreadsheet. That approach doesn’t help a recruiter screen 300 resumes faster or help a payroll manager catch compliance errors before they hit. It teaches concepts, not workflows.

If you’re a 200-person company that has invested in closing the AI skills gap in HR, this is worth sitting with. The issue isn’t the budget you’re spending on training. It’s whether what gets taught connects to the specific tasks your team does every day. A sourcing specialist who learns to use AI for candidate outreach in the context of a live requisition retains that skill. Someone who sits through a generic “AI in the workplace” webinar does not.

What to do: Audit your current AI training against actual job tasks. If your program can’t answer “what does this help a recruiter do on Tuesday?” rebuild it around workflows, not concepts. Role-specific AI training tied to real tools produces measurably better adoption. For a practical framework on where AI changes day-to-day HR work, AI in human resource management covers the functional areas worth prioritizing in 2026.

OpenAI’s Agents SDK Gets Enterprise-Grade Safety Controls

OpenAI updated its Agents SDK on April 15 with two major additions: native sandbox execution and a long-horizon model harness. The sandbox isolates agent workspaces so an agent can access only the files and code it needs for a specific task, blocking lateral movement within broader systems. The long-horizon harness handles complex multi-step tasks across extended timeframes, with configurable memory and externalized agent state so runs survive a container failure. Available to all API customers at standard pricing, Python first. (Source: TechCrunch)

For HR and ops teams evaluating AI agents for HR workflows, the sandbox directly addresses the biggest enterprise objection to deployment: what happens when an agent goes wrong? An agent processing payroll data in a sandboxed environment can’t accidentally reach into your employee records database or touch credentials it shouldn’t see. That’s not a minor technical footnote. It’s the thing that gets enterprise security teams to say yes instead of no. The AI readiness training gap isn’t only about skills — infrastructure that makes deployment safe enough to actually use matters just as much.

Salesforce Bets $50M on a Unified Agentic AI Marketplace

At TDX 2026 in San Francisco (April 15-16), Salesforce unveiled AgentExchange: a single marketplace merging AppExchange, Slack Marketplace, and the Agentforce ecosystem. More than 10,000 Salesforce apps, 2,600 Slack apps, and 1,000+ pre-built agents now live in one place. Slack AI agents grew 300% since January 2026. Salesforce backed the launch with a $50M builder initiative for partners, plus a new GoToMarket App handling private offers, unified billing, and automated provisioning. (Source: Computer Weekly | Source: Salesforce Ben)

For companies already in the Salesforce ecosystem, this simplifies a fragmented evaluation process. You no longer need to separately assess Slack apps and Salesforce apps for the same workflow. For HR teams using Salesforce for pipeline or compensation management, pre-built agents that bridge CRM data and HR workflows are now a marketplace search away. That matters when your constraint isn’t budget. It’s the bandwidth to build and test custom agents from scratch. The top AI tools for HR increasingly live in connected ecosystems rather than standalone products, and AgentExchange is the clearest bet yet that the major platforms agree.

Vietnam Becomes ASEAN’s First Nation With a Binding AI Law

Vietnam’s standalone AI law, enacted in December 2025 and effective March 1, 2026, makes it the first country in Southeast Asia to codify AI obligations into national law. Key requirements: AI systems interacting with humans must disclose their artificial nature; generated audio, image, or video content requires machine-readable markings; companies operating AI systems in Vietnam before the law’s effective date have 12 months to bring existing deployments into compliance. (Source: IAPP | Source: Baker McKenzie)

If your company hires in Vietnam or runs AI tools for Vietnam-based employees — AI screening, AI-assisted performance reviews, onboarding chatbots — you’re operating under a regulatory framework right now. The grace period runs until approximately March 2027. Use it. The labeling requirements will need coordination across product, legal, and HR before enforcement kicks in. Vietnam is widely expected to serve as a template for other ASEAN nations, so if you’re scaling in Southeast Asia, this is the model to watch.

• Nature study: humans still outperform AI agents on complex research. A landmark Nature paper found human scientists significantly beat the best AI agents on complex, open-ended scientific tasks. The implication for enterprise AI strategy: AI accelerates research, it doesn’t replace the researcher. Design your AI-human workflows accordingly, not the other way around. (Source: Nature)
• Factory raises $150M at $1.5B to build full-stack engineering agents. Factory builds AI agents that handle code review, testing, deployment, and documentation autonomously for enterprise engineering teams. Not code completion. Full software development cycle management. (Source: TechCrunch)

The Docebo numbers are a reminder that access to AI tools and actual AI adoption are two different things. If your HR team is sitting on tools they can’t apply to daily work, the fix is workflow-specific training and a stack built for it. Asanify’s HRMS is built for AI-first HR operations — if you’re evaluating where AI can actually move the needle on your team’s daily work, it’s worth exploring.

Frequently Asked Questions About the AI Readiness Training Gap

Why can’t employees apply AI training to their actual work?

The AI readiness training gap comes down to program design, not employee capability. Most enterprise AI training is generic, covering concepts like how LLMs work rather than tying to specific job workflows. Docebo’s 2026 research found that 60% of employees say training programs aren’t designed with people like them in mind, and 56% are too buried in manual pre-AI tasks to have time to apply what they’ve learned even when training is available.

What should HR leaders do to close the AI readiness training gap?

Rebuild AI training around job-specific workflows rather than generic AI literacy. A recruiter’s training should cover AI screening tools in the context of live requisitions. A payroll manager’s training should cover AI-assisted compliance checks against actual pay runs. Generic AI awareness programs produce low adoption rates. Role-specific, workflow-embedded training produces measurable behavior change and faster time-to-competency.

Does Vietnam’s new AI law affect companies hiring remotely in Southeast Asia?

Yes. Vietnam’s AI law, effective March 1, 2026, applies to any AI system used in interactions with Vietnam-based individuals — including AI hiring tools, onboarding chatbots, and AI-driven performance systems. Companies have until approximately March 2027 to bring existing AI deployments into compliance. If you hire in Vietnam or use AI tools to manage Vietnam-based employees, start the compliance conversation with your legal and HR teams now.

Not to be considered as tax, legal, financial or HR advice. Regulations change over time so please consult a lawyer, accountant  or Labour Law  expert for specific guidance.