Software Engineers Can’t “Just Code” Anymore, But Not for the Reason You Think
AI coding agents are getting very good at the parts of the job that used to be a moat: scaffolding, boilerplate, tests, migrations, refactors, even decent first-pass architecture.
So the bottleneck shifts.
Not to “being charismatic” or “winning meetings”.
To reducing ambiguity.
AI is commoditising typing, not engineering
The spicy claim is “communication is more important than code now.”
The truer claim is this: code is easier to scale, clarity is still expensive.
AI amplifies whatever you feed it. If your spec is vague or missing constraints, you’ll get confident output that can be wrong in subtle ways. The dangerous failures are the ones that look shippable until production, security review, or the next engineer tries to extend it.
So the real differentiator isn’t the ability to prompt well.
It’s the ability to make the problem crisp.
The real job is translating fuzz into something buildable
Most tickets are not requirements. They’re a starting point.
Engineers end up doing work that rarely shows up in sprint charts:
- Asking questions that reveal hidden assumptions
- Surfacing trade-offs the business didn’t realise it was making
- Cutting scope without starting a war
- Choosing defaults when nobody specified anything
- Defining what “done” actually means
That’s why some people say “you could never neglect soft skills.” In any real team, they’re right.
But there’s a valid fear in the backlash too.
When “soft skills” become the main thing, it’s usually a smell
Some organisations worship communication as a substitute for competence:
- The best talkers get promoted
- Meetings multiply
- “Alignment” replaces outcomes
- Everyone is busy, nobody ships
In those places, “soft skills” becomes a shield for mediocrity and a tool for politics.
AI hype can make it worse. It feeds the corporate fantasy that work that used to take weeks should now take hours, and if you push back you’re “resistant” or “not strategic”. Morale dies fast in that environment.
If that’s your company, the problem isn’t your soft skills.
It’s your incentives.
“Brilliant jerk” vs “friendly average” is the wrong argument
The thread also hits the classic debate: output matters, so tolerate the difficult genius.
Sometimes that works, but the impact of personality scales with coupling. If the work requires coordination, the “jerk tax” becomes real: friction, gatekeeping, rewrites, slow decisions, people avoiding collaboration.
AI makes this more visible because it increases the speed teams can sprint in the wrong direction. Coordination becomes the limiter, so bad collaboration hurts more.
Soft skills are not politics, they’re engineering hygiene
The most useful “soft skills” are operational:
- Making decisions explicit so the team doesn’t relive arguments
- Explaining trade-offs without ego
- Calling out risk early without panic
- Setting expectations so nobody is surprised late
- Asking the uncomfortable question that prevents wasted work
That’s not corporate fluff. That’s how you keep projects from quietly rotting.
What to do in 2026
If you want practical leverage:
- Be ruthless about clarity
If requirements are vague, make them less vague. If nobody can answer, propose defaults and get explicit sign-off. - Measure what matters
Use numbers to drive decisions: latency, error rates, churn, incident frequency. Metrics end vibe debates. - Treat scope like a budget
Every “small request” has a cost in complexity and risk. Make the cost visible early. - Write for your future self
Short docs, clear decisions, assumptions stated, trade-offs logged. This becomes your multiplier. - Use AI to accelerate, not to cosplay
If you can’t explain it, you don’t own it.
Conclusion
Hard skills still matter. Soft skills still matter. AI raises the floor on code and raises the ceiling on impact.
Your edge in 2026 isn’t typing faster.
It’s turning unclear reality into a buildable plan, shipping something durable, and doing it in a way that doesn’t burn the team down.
Member discussion