When Expertise Becomes Weight
Why experienced developers can adapt more slowly to agentic coding, and how expertise turns into fixation.
Key takeaways
- Experienced developers can adapt to agentic coding more slowly because their fast pattern-recognition loop (recognize then implement) fires before they notice and blocks the design-delegate-verify loop.
- The Einstellung effect explains the fixation: once a known solution comes to mind, skilled people stay attached to it even when a cheaper or safer path exists.
- The tell is a single sentence, "this has to be done this way"; the new skill is pausing to check whether that judgment is still current.
- Over-design is not always wrong, but the automatic reaction to abstract before analyzing the actual requirement (e.g., a direct Prisma query versus a filter-strategy framework) is the trap.
- A one-week practice helps: for every task you would "just write yourself," write the intent in three sentences, ask AI for two approaches, and compare against your default.
A senior developer with 10 years of experience receives an agentic coding tool and adapts more slowly than a junior developer who joined three months ago. This is not a joke. It is a repeated pattern.
The issue is not lack of ability. It is that ability has become an automatic loop.
The Expert's Paradox
The strength of experienced developers is pattern recognition. They see a problem, map it to a past solution, and start implementing before a junior developer has finished reading the ticket.
That strength can invert in agentic coding.
Traditional development rewarded the loop recognize pattern -> implement directly. Agentic development rewards design intent -> delegate -> verify. The first loop is so trained that it fires before you notice it.
A Familiar Example
For a user filtering API, an experienced developer may start by designing reusable filter strategies, composite filters, and a future-proof abstraction. That can be reasonable if 20 entities need the same behavior.
But if the current requirement is "filter users by name, role, and signup date", a direct Prisma query is often better:
export async function getFilteredUsers(filters: {
name?: string
role?: string
since?: Date
}) {
const where: Prisma.UserWhereInput = {}
if (filters.name) where.name = { contains: filters.name, mode: 'insensitive' }
if (filters.role) where.role = filters.role
if (filters.since) where.createdAt = { gte: filters.since }
return prisma.user.findMany({ where, orderBy: { createdAt: 'desc' } })
}The real cost of over-design
The abstract version is not automatically bad. The problem is the automatic reaction: "filtering should always be built this way" before the current requirement has been analyzed.
The Einstellung Effect
Psychology has a name for this: the Einstellung effect. Once a known solution comes to mind, people stay attached to it even when a better solution exists. Skilled people can be more vulnerable because their known solutions are fast and usually useful.
| Situation | Automatic expert reaction | Agentic alternative |
|---|---|---|
| REST API design | "This pattern always goes here" | Describe intent and ask for several patterns |
| Error handling | Copy an old pattern and modify it | Specify scenarios and ask for context-fit handling |
| DB migration | Apply the familiar sequence | State constraints and ask for safe paths |
| Performance work | Apply a favorite optimization | Give profiling data and ask for hypotheses |
| Code review | Reject code because it uses an unfamiliar pattern | Check behavior, risk, and fit to intent |
The alternative is not always better. The point is that the automatic reaction prevents you from trying it.
Where Adaptation Gets Slower
Observation-based model
The following ranges are not industry statistics. They are an explanatory pattern observed across teams adopting agentic coding tools.
| Experience range | Common adaptation range | Main barrier |
|---|---|---|
| Junior, 0-2 years | 1-2 weeks | Harder to verify outputs |
| Mid-level, 3-5 years | 3-4 weeks | Some patterns are fixed, but flexibility remains |
| Senior, 6-10 years | 6-8 weeks | Strong pattern fixation tied to identity |
| Staff, 10+ years | 8-10 weeks | Strong fixation plus organizational influence |
The message is not "senior developers are worse." The message is "senior developers have more to unlearn."
The Sentence to Notice
Most friction begins with one sentence:
"This has to be done this way."
Sometimes that sentence protects the system. Sometimes it blocks a cheaper and safer path. The new skill is not abandoning judgment. It is pausing long enough to ask whether the judgment is current.
Practice
For one week, mark every task where you think "I can just write this myself." For each one, run a small experiment:
- Write the intent in three sentences.
- Ask AI for two approaches.
- Compare one AI approach against your default approach.
- Adopt, reject, or combine them based on system fit.
The goal is not to use AI every time. The goal is to break the automatic path.
Next
Read The Illusion That I Must Write Everything to examine the identity layer behind this habit.