I had a coaching conversation recently that stuck with me. A developer, ten-plus years in, told me something to the effect that he could write the same function in about the same time as the AI now, and he said it like a confession, not a brag. That’s the second fault line in this series, and it’s a quieter one than title loss. Nobody strips your badge in a meeting. The thing that erodes is the private certainty that your hands-on mastery is the reason you’re in the room.
Here’s the part I think gets missed: the fear isn’t really “AI can code.” It’s “the thing I spent a decade getting good at is no longer the scarce thing.” That’s a different injury, and it needs a different response than the one we used for title loss.
The Leveling Effect Is Real, and It’s Not the Whole Story
There’s a useful piece of research modeling exactly how generative AI redistributes skill value, and it lands on two very different outcomes depending on the task. In some studies, customer service agents using an AI assistant, consultants using ChatGPT, and even programmers with early Copilot, the tool mostly closed the gap between novices and experts, letting less experienced people perform at a near-expert level. Researchers call this a leveling effect, and it’s the mechanism behind the fear you’re sensing: if a newer hire can match your output with the tool, what was all that deliberate practice actually building?
But the same research found the opposite can happen too, a multiplier effect, where AI amplifies the gap between experienced and inexperienced people instead of closing it. The deciding factor isn’t the tool. It’s the task. When the AI’s output is easy to evaluate and barely needs editing, leveling wins; anyone can accept a good-enough answer. When the output requires real judgment to assess, tweak, and catch subtle errors, expertise still wins, because knowing when the AI is wrong is itself the skill. One sales study even found this directly: agents using an AI system that screened leads still saw top performers outsell bottom performers by nearly three to one because the top agents knew how to handle the parts the AI couldn’t touch.
This reframes the whole anxiety. The question isn’t “Will AI level my skill.” It’s “Is my skill mostly about producing output or mostly about judging it.” If it’s the former, you’re right to be nervous. If it’s the latter, you may be sitting on more leverage than you think.
The Quiet Danger Nobody’s Naming: Skill Erosion From the Inside
There’s a second, subtler risk buried in this same research, and it’s the one I think deserves more airtime in coaching conversations. If a leveling tool is doing your evaluation and editing for you, not just your first draft, you stop getting the reps that built the skill in the first place. One case study found a company’s accounting staff became so dependent on an automated tool that when the software was swapped out, the team literally couldn’t perform the underlying process anymore. Researchers call that skill erosion, and it’s a slow leak, not a sudden loss.
This is where I think the T-shaped framing becomes genuinely practical instead of just a nice diagram from the Scrum Guide. A T-shaped person carries deep expertise (the vertical bar) and broad working fluency across adjacent domains (the horizontal bar), originally so a team could swarm around bottlenecks instead of stalling on handoffs. Under AI pressure, that same shape is doing something new: the vertical bar is what’s shrinking in relative importance as AI absorbs deep-specialist output, while the horizontal bar, judgment, context, and cross-domain fluency are what’s left standing. Some people are already describing where this goes next: from I-shaped (narrow depth) to T-shaped (depth plus breadth) to V-shaped or M-shaped builders who can move end-to-end across a problem instead of staying in one lane. I don’t think everyone needs to become an M-shaped generalist. But I do think the horizontal bar is no longer optional — it’s becoming the part of your skill set that AI can’t quietly absorb while you’re not looking.
What Actually Fills the Horizontal Bar Now
If the deep vertical skill is compressing, the practical question is what to build instead. Recent analysis of AI-era engineering work gets specific: the horizontal bar is increasingly made of AI literacy, understanding how a model reasons and where it typically fails, not just how to prompt it, plus context engineering, data fluency, product empathy, and the ethical judgment to know when a “good enough” AI answer isn’t actually good enough. Notice what that list has in common: none of it is a task AI performs. It’s the layer of judgment that decides whether the AI’s output should be trusted, shipped, or thrown out.
This lines up with something Anders Ericsson’s deliberate practice research has said for decades, just aimed at a new target: mastery comes from focused, feedback-driven repetition just beyond your current ability, guided by a coach who can tell you what to work on next. The mistake would be assuming deliberate practice no longer applies because AI writes the first draft. It applies to a different rep now, the rep of catching what the AI got subtly wrong, again and again, until that judgment becomes your new deep skill.
The Psychological Buffer That Actually Shows Up in the Data
I want to bring in one more finding because it’s the closest thing I’ve seen to hard evidence for something coaches say all the time on faith. A study of 444 professionals found that people with strong “career adaptability” a measurable mix of future orientation, sense of control, curiosity, and confidence, reported significantly lower AI anxiety than people without it, and the effect showed up in a clean, three-tier pattern: low adaptability, highest anxiety; high adaptability, lowest anxiety. The mechanism is worth sitting with: adaptability didn’t just reduce anxiety directly; it worked by strengthening people’s core belief in their own competence and control, and that self-belief is what actually did the anxiety-reducing work.
That’s a mechanism you can coach toward. You can’t hand someone confidence directly. But you can put them in front of small, winnable challenges that build a track record of “I adapted to that,” and the adaptability itself becomes the thing that protects them the next time the ground shifts.
What to Actually Do With This Monday Morning
- Ask people what part of their job requires judging output, not producing it; that’s the part least exposed to leveling, and most worth naming out loud as valuable.
- Watch for over-reliance quietly replacing the reps that built someone’s skill in the first place. If AI is doing the evaluation too, not just the first draft, the skill is eroding even if the output looks fine.
- Stop treating the horizontal bar as a nice-to-have. AI literacy, context judgment, and product empathy are the new deep skills, not side skills.
- Give people small, winnable adaptation challenges. Career adaptability isn’t a trait people either have or don’t; it’s built through reps of successfully handling change, the same way any other skill is.
Where I Land, For Now
The deep, hands-on mastery that got you the title was never actually the whole point; it was always a proxy for judgment, for knowing what “good” looks like well enough to catch what’s wrong. AI is taking the proxy. It hasn’t taken the judgment. The people who come out of this fine aren’t the ones who out-code the machine. They’re the ones who get faster at knowing when it’s wrong.
The next post in this series goes after the third fault line: what happens when it’s not your title or your skill that is shaken, but your sense that you’re still the one actually building something. That one, I think, is the deepest cut of the three.
- Deskilling and upskilling with generative AI systems – by K Crowston · Cited by 61 — In this paper we focus on a long-standing concern about the impact of …
- ‘T’-shaped People Make Scrum Teams More Efficient – T-shaped individuals tend to possess both deep skills in their particular area of expertise as well …
- Liam Darmody – Most teams hire for T-Shaped talent. – Most teams hire for T-Shaped talent. The best ones grow V and M-Shaped builders. Most of us know the…
- Understanding Scrum Development Teams: T-Shaped vs. I … – Specialized Skills: In a Scrum team, a T-shaped member is someone with a strong primary skill or are…
- Is the T-Shaped Developer the Future of AI-Led Software … – T-shaped teams form what can be described as a blend of “generalized specialists” and “specialized g…
- T-shaped Skills and Swarming Make for Flexible Scrum and … – T-shaped skills is a metaphor used to describe a person with deep vertical skills in a specialized a…
- Richard Hughes-Jones’ Post – Deliberate practice is a focused and systematic approach to improving skills and achieving expertise…
- Deliberate Practice – The Ultimate Key to Skill Mastery – Deliberate practice is an approach to learning that leads to mastery in any skill or sphere of knowl…
- The impact of career adapt-abilities on AI anxiety among … – by X Wang · 2026 · Cited by 3 — AI anxiety has been shown to significantly influence learning engage…
Leave a Reply