Directing Intelligence Digest | October 13, 2026 (Experimental)
Today’s through-line is simple... and human.
AI is expanding what people can do at work and in school. The data says so. The X debate says so. And still, the parts that make us us (judgment, empathy, verification, deliberate learning) are not going away. They are becoming the work we need to do.
Curious professionals, leaders, educators: this is your digest.
1. German IT apprentices: more tasks, not less understanding (at least right away)
What happened. A Stanford Digital Economy Lab working paper (PDF dated Oct 7, 2026) reports a field experiment with 673 final-year IT apprentices across twelve German vocational schools. Some got ChatGPT Pro. Others worked without internet or AI. Tasks sat inside the apprenticeship curriculum... and deliberately beyond it.
Why it matters for people. Measured productivity (share of maximum points) rose by 26 to 44 percentage points across post-treatment tasks. On beyond-curriculum work, gains were 26 to 31 percentage points. Immediate comprehension did not fall; on some checks it improved. Effects were generally larger for apprentices with above-median AI literacy.
So early-career, middle-skilled people may reach further with AI beside them. That is agency... if we also teach them to verify, explain, and own the result. A higher assisted score is not the same as durable mastery months later. The paper is careful about that. We should be too.
What to watch. Do employers redesign junior roles around verified understanding, or only around faster tickets? Do vocational systems teach AI literacy as a core skill, not a side hobby?
Read it: Task Expansion with Generative AI (PDF) · Stanford project page
2. Chile’s Kai: information moved college choices more than empathy alone
What happened. Gregory Elacqua flagged a new IDB / Elige Educar nationwide RCT in Chile (author post: N≈42,000; CEPR abstract: N=41,000). High school seniors got WhatsApp counseling from either the AI counselor Kai or trained human counselors, aimed in part at teaching careers and education majors.
Why it matters for people. Among students with baseline interest in education / teaching, Kai significantly raised first-ranked education majors and the share of education majors in choice sets. Human-arm estimates were smaller and not statistically significant. Kai matched or outperformed roughly two-thirds of human counselors.
How? Text analysis: Kai leaned factual, faster, more coherent. Humans scored higher on motivation and empathy. Elacqua’s punchline: Kai won on information, humans on empathy... and information mattered more for moving choices.
That is a human story, not a robot victory lap. Scale and consistent facts can open doors for students who never get a counselor. Empathy still matters for dignity. The future is probably both... designed on purpose.
What to watch. Hybrid counseling. Whether other countries replicate the info-vs-empathy split. Long-run teacher pipeline outcomes, not just application clicks.
Read it: IDB paper listing · CEPR DP21658 abstract
On X: Gregory Elacqua
3. UK CEOs: AI shows up in productivity... skills show up as the bottleneck
What happened. EY’s UK newsroom (Oct 12, 2026) reports results from 100 UK CEOs in the EY-Parthenon CEO Outlook work.
Why it matters for people. 50% put AI-enabled tools in their top three drivers of productivity gains over the past year. Only 14% have comprehensive, real-time visibility of AI costs. 74% expect skills shortages to beat access to capital as a growth barrier within three years. 71% say AI will hit roles, skills, and how work is organised more than headcount. Nearly half (47%) say their organisation is not developing skills fast enough.
In plain English: the chart that keeps people up at night may not be layoffs. It may be whether your role still fits you... and whether anyone invests in helping you grow into the new version.
What to watch. ROI dashboards that include skills investment. Entry-level pathways that still teach judgment. Leaders who reinvest productivity into people, not only margin.
Read it: EY UK: AI drives productivity gains, but barriers remain
4. Software’s 25-50% reality check... and the verification leap that is not here yet
What happened. Zoho’s Sridhar Vembu posted an end-to-end take on AI in software development.
Why it matters for people. He puts today’s AI-led software productivity at at best 25% to 50% on a finished-product basis. Prototyping is much faster. Shipping still needs a lot of human work. A 10x / 100x leap could come, he argues, via auto-formalization and Lean-style verification loops... but it has not happened yet. Meanwhile server costs and token spend are climbing, which he frames as pressure on software jobs and a call for reinvention.
For people who write and ship software: speed at the blank page is not the same as responsibility for the finished system. Verification, integration, and ownership remain human work until proven otherwise.
What to watch. Cost per shipped feature. Whether verification tools change junior vs senior work. Teams that reinvent roles before the leap arrives... not after.
On X: Sridhar Vembu
5. Learning design: novices still need explicit guidance
What happened. Education researcher Carl Hendrick: minimally guided instruction does not work with AI tutors either. Socratic questioning models frustrate students when they withhold the explicit guidance novices need.
Why it matters for people. A confused learner is not a deeper learner. Dignity in education means scaffolds matched to expertise. Agents that only ask questions can look wise... and leave beginners stuck.
What to watch. Adaptive tutors that know when to explain. Schools that refuse one-size-fits-all “Socratic-only” AI.
On X: Carl Hendrick
6. Knowledge work without offloading your brain
What happened. Princeton’s Arvind Narayanan, after The Ezra Klein Show, described preparing with an agent that has read his own output (books, essays, talks, transcripts, posts, notes) on the order of a million words to surface pithy talking points. He calls it beneficial knowledge-work AI that does not involve cognitive offloading.
Why it matters for people. The best tools can amplify your judgment and voice. You stay the editor-in-chief. That is a different moral story than paste-and-ship slop.
He also points back to the AI as Normal Technology frame: gradual economic impacts, uneven labor effects, changing knowledge work. Agency stays with institutions and people who choose how much power to grant tools.
What to watch. Personal knowledge systems that keep humans accountable. Workplaces that reward judgment over volume.
On X: Arvind Narayanan
Read / listen: Ezra Klein Show episode page · AI as Normal Technology (related essay)
By the numbers
673 German IT apprentices in the Stanford field experiment; productivity gains of 26-44 pp overall and 26-31 pp on beyond-curriculum tasks. Source
Chile counseling RCT: about 41,000-42,000 students; Kai matched or beat roughly two-thirds of human counselors on outcomes for the target group. IDB · Elacqua on X
50% of surveyed UK CEOs put AI among top three productivity drivers; 74% expect skills shortages to outrank capital as a growth barrier; 71% say AI changes roles/skills more than headcount. EY
Software end-to-end AI productivity today: at best 25-50%, per Vembu; 10x/100x not yet. X
Worth a listen
Lenny’s Podcast with Josh Woodward (Google Labs): why companies need to operate like a lab team... and why unlearning rate may be a rising skill. Episode · YouTube (published Oct 11, 2026)
The AI Daily Brief: How to Go from AI Power User to AI Leader on building AI champions networks so good practice does not die with a few power users. Episode (published Oct 12, 2026)
Tools expand the reach of hands and minds. People still decide what is true enough to ship, kind enough to say, and worth learning on purpose.
Otherwise it is not us. See you tomorrow.
Sources
https://digitaleconomy.stanford.edu/app/uploads/2026/10/AI_Task_Expansion_07102026.pdf
https://publications.iadb.org/en/scaling-counseling-ai-evidence-nationwide-experiment
