Hi, I'm Charles Costa, researcher, strategist, and founder of Lexora Labs. Each week I dig into AI, knowledge management, and future-of-work research so you don't have to.
EDITOR'S NOTE
Last week we mapped why AI adoption feels so uneven. This week we flip the lens: once AI is working, what's left for the humans? Five reads land on the same fault line, AI moves fast but flattens nuance and judgment, and the companies pulling ahead aren't automating the most, they're deciding precisely where a human still belongs. If you're planning headcount for 2027, figure out which judgment calls are irreplaceable, and start pricing them that way.
THIS WEEK'S READING:
Employers who laid off workers citing AI are already starting to regret it — CNBC Commonwealth Bank of Australia and IBM both cut roles for AI this year, then reversed course, CBA called its cuts an "error," and IBM tripled entry-level hiring after its AI could resolve only 94% of HR requests alone.
Why It Matters: AI is the new hire who nailed the interview but never sat in on a client call, it fails exactly where tribal knowledge matters, and fails confidently. Our own research on AI in customer service found the same pattern: the companies getting this right use AI to clear routine volume so humans stay on the complex cases.
That failure is baked into how these systems are built and measured in the first place.
Why AI systems fail at scale and what you should measure instead of model accuracy — CIO Most enterprise AI failures trace back to architecture, not model quality: systems buckle under load, prediction quality erodes as models drift, and even good systems need a containment plan for what slips through.
Why It Matters: Accuracy isn't relevance, a model can correctly interpret a product requirements document (PRD) from July and be dead wrong in August. The fix isn't a smarter model, it's training your team to spot a stale answer, and favoring specialized models over one system doing everything.
If the system itself is the risk, the obvious fix is adding a human to check it. The data on that is messier than it sounds.
When combinations of humans and AI are useful: A systematic review and meta-analysis — Nature Human Behaviour Analyzing 106 experiments, researchers found human-AI combinations typically underperform the stronger of the two alone, with gains in creative tasks and losses in decision-making, and over-reliance on AI drives the losses.
Why It Matters: Not an argument against AI, an argument against generic deployment. Our own research found the same thing: AI adds the most value tailored to a company's knowledge base, not dropped in off the shelf.
That decision-vs-creation split isn't academic. One widely-read UX voice just showed how bad the decision-making losses get.
When Humans Add Negative Value: AI Alone vs. Human-AI Synergy — Jakob Nielsen, UX Tigers Nielsen argues a human reviewer doesn't uniformly help, on convergent tasks like forecasting, humans work from gut instinct while AI follows a consistent methodology, so the override does more harm than good; the exception is creative work, where collaboration adds value.
Why It Matters: The dividing line isn't human vs. AI, it's convergent vs. creative. Before adding a human review step, ask whether the task has a right answer a model can calculate, or needs judgment the AI can't supply.
Knowing which tasks need a human is one thing. The labor market is already pricing that judgment into salaries.
2026 Global AI Jobs Barometer — PwC PwC's global data shows the most AI-exposed junior roles are seven times more likely to demand senior skills like leadership, while the companies posting the biggest AI productivity gains aren't cutting headcount fastest, they're growing headcount and wages faster than less-exposed companies.
Why It Matters: This undercuts the "AI kills the entry-level rung" narrative, it's re-pricing junior roles around leadership and judgment since AI absorbs the grunt work. The companies winning with AI use it to unlock growth, not just cut costs.
Every read this week lands on the same line, not human vs. AI, but convergent vs. creative, routine vs. judgment-heavy. The founders getting this right aren't cutting headcount and hoping the model catches everything, they're deciding, deliberately, where a person still belongs, and pricing their teams around that judgment.
Have a take on any of this? Reply directly, I read every response.
Until next Tuesday,
Charles Costa, MLIS