State of the Future of Work
What the market is starting to verify
July 2026 edition
Each edition, we read what practitioners and researchers are actually saying about AI and work — and hold it against one question: what does it mean for how capability gets proven? No predictions dressed as certainty. Three pieces stood out this month, and they converge on the same point from three very different directions.
From the platform trenches
ZTA: Zero Token Architecture
Kelsey Hightower, PlatformCon 2026 Live Day New York
Hightower's provocation is in the name: an architecture built around not burning tokens. His argument is that the fundamentals — understanding your system, knowing what problem you're solving, designing before generating — still decide outcomes, and that reaching for the model is a choice to be justified, not a default.
Why it matters for provenance: if the loudest failure mode of this era is mistaking AI consumption for AI capability, Hightower just named the antidote. The professionals worth hiring are the ones whose record shows judgment about when and why they engaged the tool — not the ones with the biggest token bill. Usage volume was never the signal. The decision quality around the usage is.
From inside the labs
The unreasonable effectiveness of HTML
Thariq Shihipar, Claude Code team, claude.com
A Claude Code engineer explains why he stopped having the agent write Markdown and started demanding HTML: richer artifacts are easier to actually read, interrogate, and share — so he reviews more of what the model produces instead of skimming past it. His words: it keeps him "in the loop" as the agent takes on more.
Why it matters for provenance: notice what the differentiating skill is here. Not prompting. Not model choice. It's verification discipline — deliberately shaping the workflow so the human can check the work. That's exactly the behavior that separates professionals in captured sessions: pushback, requests for evidence, output formats chosen for reviewability. It's invisible on a resume and unmistakable in the record of the work itself.
From the market's most-watched voice
Expertise, redefined: deep domain knowledge plus AI as a tool
Jensen Huang, NVIDIA — recurring theme across recent talks and interviews
Huang's consistent framing of the future of expertise has three parts: pick a domain and go deep, reason from first principles rather than copying, and treat fluency with AI as a non-negotiable skill — what he calls accelerated mastery. He adds a fourth, quieter one: the strongest people combine technical depth with the empathy to see what's unsaid.
Why it matters for provenance: none of those four things appear on a degree or a title. Every one of them appears in the record of real work — how someone frames a problem, what they verify, where their domain depth shows up in the questions they ask. When the market's most influential hiring voice defines expertise in terms that credentials can't certify, the verification gap becomes the story.
The pattern this month
A platform engineer, an AI lab, and a chip CEO, all landing on the same claim: fluency is judgment applied to AI, not consumption of it. Fundamentals decide outcomes; verification is the skill; domain depth is the moat. What the market is starting to verify is how people work. What it still takes on faith is everything a resume asserts. That gap is closing from one side only — and the professionals capturing evidence of their judgment now are on the right side of it.
Your AI collaboration already generates this signal. ValueSignal captures it, scores your contribution — not the AI's — and keeps it private until you choose to share.
Open your workspace