← narwal.one/Second Brain
SecondBrain
Ask the Brain
Index/Conceptupdated Tue Oct 27 2026 08:00:00 GMT+0800 (Philippine Standard Time)

The Upskilling Bill

upskillingenablementtrainingagentic-transformationwork-primitivewhittemoreai-daily-brief

The Upskilling Bill

Nathaniel Whittemore's phrase for the enablement debt that agentic AI has made due: "I believe that the upskilling bill is coming due in a huge way."

The argument is about a change in kind, not degree:

"When AI learning was just about whether you could prompt well, maybe you could get away with not investing a ton in training your workforce. Now, on the other hand, we are talking about a fundamentally new work primitive. The way that people work is changing in a core way in many disciplines and functions — from 'I do my work' to 'I manage agents that do my work for me.' The need that that creates for training is radically heightened from the previous era of AI."

Why the prompt-era playbook doesn't clear the bill

Prompt era Agentic era
Skill = phrasing a request well Skill = specifying, delegating, supervising, accepting
Failure mode = a bad answer you can see Failure mode = an agent acting on systems you weren't watching
Blast radius = one output Blast radius = whatever the agent had keys to
Training = a course Training = supervised practice on real work

The consequence Whittemore reports from the KPMG floor is not theoretical. Multiple stories circulated of people accidentally unleashing agents on critical systems — and his framing of blame matters:

"…not because even necessarily they were doing anything wrong, but because there weren't the right guardrails or access provisioning, and these incredibly capable models with their new tenacity just didn't stay in their boxes."

So the bill has two line items, and paying only one fails:

  1. Enablement — teach people the new primitive.
  2. Provisioning — scope what agents can touch at the harness level, not by instruction (Capabilities vs Instructions (Agent Keys)).

He is explicit that training alone is insufficient — "this is not an upskilling question alone. Again, the watchword of the moment is systems and architectures" — but that without it organizations are "almost doomed" to repeat the incidents, or to over-correct and lock out the people who would generate the most value.

What enterprises are actually doing about it

From the symposium floor:

  • Building it themselves. "A lot of throwing up of the hands and saying screw it, we're just going to have to do this ourselves" — bespoke programs matched to the org and its population, rather than off-the-shelf curricula.
  • Rejecting the course format. Explicitly not "a bunch of cute video courses of the pattern of corporate trainings of yore", but "the real messy work of getting people to use these tools in new ways to do new things."
  • Solving transmission, not just instruction. The recurring design problem is moving knowledge from the parts of the organization figuring it out well to the parts that aren't.
  • Engineer → business-unit pairing as the dominant pattern. Not substitution: "no one is talking about the marketing folks replacing the engineers, but they are now talking about the 10 or 20% of the types of skills — and even more than that, mindsets — that engineers or product managers have that can become a part of the essential toolkit" for marketing, sales, back office and the rest.

That last point is the most actionable: the transferable asset is a mindset subset, not a technical curriculum. It's the same shape as the vault's Agentic Pods pattern (Uber's 30 engineers × 30 domain experts) — pairing, not retraining.

2026-10-20 — How the bill actually gets paid: champions (per How to Go from AI Power User to AI Leader (AI Daily Brief))

The "solving transmission, not just instruction" problem above now has a named mechanism in the vault. Nufar Gaspar and Whittemore argue that a structured AI Champions Network is the highest-ROI line item on the bill precisely because it is small (a few people per team), contextual (trained on their own work, certified on one real use case), and owned after certification (two sponsors; "the biggest predictor of whether this succeeds is what happens the day after"). Whittemore's addendum echoes the KPMG floor: AI training "doesn't fit comfortably into previous models of corporate upskilling" and has no agreed best practice, which is exactly why investing in the disseminating group first is the safe bet. The provisioning half reappears too — non-technical staff using agentic coding tools whose agents "escape containment pretty quickly" — and the prescription is enablement plus infrastructure, not shutdown.

Cross-references

Sources

2026-10-27 — Who is paying and who is not (per Enterprise AI Adoption and the Reshaping of Job Profiles (Research Brief 2026-10-27))

Three datapoints from the October window. Not paying: Barclays' ~5,000 cuts arrived with a productivity framing and no reskilling commitment; the Unite union called for upskilling into AI-adjacent work instead (CNBC, 2026-10-22). Paying, unevenly: SHRM's 2026 HR survey (via Workmate) finds 57% of organisations seeing more upskilling activity but only 28% saying HR leads it, and 39% with AI in HR at all; Deloitte's enterprise survey has 53% raising broad AI fluency against 48% with a formal reskilling strategy. How it gets paid: the X datapoint that 60% weekly active usage across 6,000 employees came "only when workers build and deploy their own custom agents rather than relying on top down tools" (@ai_mindsss) is the same transmission argument as the champions section above, and India's IT services firms are answering with role-based academies and certification pathways (CIEL HR via Outlook Business).