Pergeseran Fase Besar: Mengapa Saya Berhenti Mengkode (dan Mulai Mengorkestrasi) pada 2026

TL;DR: Around December 2025, a threshold was crossed. I went from writing 90% of my code manually to letting AI write 90% of it while I direct the strategy. This isn't just a productivity boost; it is a fundamental rewriting of what it means to be a software engineer. We are entering the era of "Code Actions," where stamina, not syntax, is the new bottleneck.

James here, CEO of Mercury Technology Solutions.

In my three decades of programming, I have never experienced a shift this violent or this fast.

In November 2025, my workflow was 70% manual coding and 30% AI assistance.

By December 2025, that ratio flipped: 90% Agent coding, 10% human review.

I am essentially programming in English now. It hurts the ego to tell a machine what to write in plain text, but the leverage is undeniable.

We are no longer writing lines of code; we are executing large "Code Actions."

Here is my field report on the new reality of software engineering.

1. The Reality of the "Agent Swarm" (Don't Fire Your IDE)

There is too much hype about "Agent Swarms" replacing IDEs.

Do not delete your IDE. You need to watch these agents like a hawk.

The nature of bugs has changed.

• Old Bugs: Syntax errors, typos, null pointers.
• New Bugs: Subtle conceptual errors. The Agent makes a wrong assumption and runs with it confidently.

The "Junior Dev" Syndrome:

Agents today act like sloppy, hasty junior developers.

• They don't manage confusion or ask for clarification.
• They are sycophantic (they agree with your bad ideas).
• They love overcomplicating APIs and bloating abstractions.
• They forget to clean up dead code.

You will often see an Agent write a brittle 1,000-line monster. You have to step in and say, "Couldn't we just do this in 100 lines?" and it will immediately correct itself.

My current setup: AI sessions (Ghostty) on the left for generation, IDE on the right for discrimination and review.

2. Tenacity: The New Superpower

The most "Feel the AGI" moment isn't the code itself; it's the Stamina.

Agents never get tired. They never get demoralized.

I watched an agent struggle with a problem for 30 minutes, trying approach after approach, until it emerged victorious. A human would have given up to get coffee.

Stamina is a core bottleneck to work. AI has removed that bottleneck.

3. Leverage: Imperative \~ Declarative

The real magic happens when you stop telling the AI how to code and start telling it what success looks like.

• Don't say: "Write a function that iterates through X..."
• Do say: "Here is the success criteria. Write the tests first. Keep looping until the tests pass."

This shift from Imperative (giving orders) to Declarative (defining goals) is how you gain massive leverage. You become the architect; the AI is the builder.

4. The "Speedup" Paradox (Expansion > Velocity)

Am I faster? Yes.

But the main effect isn't speed; it's Expansion.

1. Scope Expansion: I build things I previously wouldn't have bothered with because the ROI wasn't there.
2. Skill Expansion: I tackle domains where I previously lacked the technical skills.

The "Speedup" is actually an explosion of capability.

5. The Risks: Atrophy and The Slopacolypse

Skill Atrophy:

I can already feel my ability to write code manually fading. The brain separates "Generation" (writing) from "Discrimination" (reading). I can still review code perfectly, but the muscle memory for syntax is dying.

The Slopacolypse of 2026:

Brace yourselves. GitHub, Substack, Arxiv, and X are about to be flooded with low-quality, AI-generated "slop." We are entering an era of "Productivity Theater," where volume completely disconnects from value.

The Big Questions for 2026

As we metabolize this new capability, four questions keep me up at night:

1. The 100x Engineer: What happens to the ratio between the average engineer and the best? Armed with AI, the gap between the mean and the max might grow exponentially.
2. Generalists vs. Specialists: LLMs are great at the "Micro" (filling in blanks). Humans are needed for the "Macro" (Grand Strategy). Does this mean the era of the Generalist (me) has truly arrived?
3. The Gamification of Code: Does coding feel like playing StarCraft (managing units) or Factorio (optimizing systems) now?
4. The Digital Bottleneck: How much of society was held back simply because we didn't have enough people to write the code? We are about to find out.

Conclusion: The High-Energy Year

We crossed a threshold of coherence in late 2025.

The intelligence is now ahead of the integration. The tools, workflows, systemic design and organizational structures haven't caught up yet.

2026 will be a chaotic, high-energy year as the industry figures out how to wield this new weapon.

Are you still coding, or are you orchestrating?

Mercury Technology Solutions: Accelerate Digitality.