30% of Code Has Been Taken Over by AI — What Can Engineers Still Do?
On the evening of April 29 (U.S. time), during Meta’s LlamaCon 2025, Mark Zuckerberg (CEO of Meta) and Satya Nadella (CEO of Microsoft) appeared on stage together for the first time — and they made the same bold declaration:
“Engineers are no longer people who write code.
They are commanders of AI teams.”
GitHub Copilot has evolved into a “prototype agent,” capable of generating pull requests and managing workflows independently.
Inside Meta, each developer is now paired with several small AI “apprentices” working under their direction.
Their shared conclusion was simple:
“We’re not eliminating engineers — but we are completely recompiling their roles.”
This article breaks down that conversation through key questions:
- After AI writes 30% of the code, what’s left for engineers to do?
- What does it actually mean to “lead an AI squad”?
By the end, you’ll have a blueprint for the next-generation engineer.
Part 1｜AI Hasn’t Taken Jobs — It’s Changing What “Work” Means
“Inside our company, up to 30% of new code is now written by AI.”
— Satya Nadella, LlamaCon 2025
Nadella’s words echoed through the conference hall.
He continued:
“Copilot no longer just completes code. It can handle prototype tasks end-to-end — generating PRs, orchestrating code workflows.
This isn’t assistance. It’s takeover.”
The room fell silent for a second — then erupted in applause.
But the real signal here isn’t the 30%, it’s the structure of what’s changing:
AI isn’t stealing jobs — it’s quietly rewriting the definition of work itself.
The “Deconstruction” of the Engineer’s Actions
Zuckerberg followed with a story from inside Meta:
“We have a team that works entirely through ‘agent-based collaboration.’
Developers break tasks into pieces, delegate each to smaller models — code generation, debugging, testing, committing.
One human engineer only steps in at the end for structural refinement and quality review.
AI handles about 80% of the total workload.”
He called this phenomenon “the decoupling of engineering actions.”
AI is stripping away repetitive and standardized parts of a developer’s day.
Once, engineers were end-to-end builders:
- Write requirements
- Design architecture
- Implement logic
- Debug and deploy
Now, engineers are task orchestrators and system maintainers.
It’s no longer about typing speed or syntax memory — but about intent decomposition and resource orchestration.
As Zuckerberg put it:
“AI will do what you tell it to do.
The question is not how well it writes, but how clearly you tell it what to do.”
AI Doesn’t Just Execute — It Redefines Execution Itself
Nadella made a subtle but powerful point:
“If you still see AI as a productivity tool, you’re in phase one.
The real turning point is when AI doesn’t just accelerate old workflows — it creates new ones.”
He cited Microsoft’s internal sales workflow as an example:
Before: preparing for a client meeting meant manually searching, writing briefings, sending emails.
Now: Copilot automatically integrates CRM data, company updates, and team communications to generate client proposals in real time.
Work output has shifted from Word documents to interactive dashboards.
The process itself has disappeared.
From “Using AI” to “AI Working — Humans Judging”
Both CEOs agreed:
- Engineers shouldn’t try to “code harder,” but to manage AI smarter.
- Developers must shift from hands-on execution to structural design.
- Human work isn’t vanishing — it’s moving up the stack, into the logic that governs AI systems.
Zuckerberg even predicted:
“By next year, half of Meta’s development will be led by AI.”
And Nadella concluded:
“We’re entering an era where for every line of code, we’ll ask:
Does a human really need to do this?”
Part 2｜Engineers as AI Orchestrators, Not Coders
“Future engineers will be like technical directors — commanding a squad of AI models from design to deployment.”
— Mark Zuckerberg, LlamaCon 2025
This was one of Zuckerberg’s most visionary statements.
Instead of old clichés like “AI productivity” or “AI collaboration,” he introduced a new archetype:
The Technical Commander
In the past, engineers were valuable because they could do everything themselves.
In the future, they’ll be valuable because they can coordinate many models to do everything together.
From Hands-On Skills to Orchestration Skills
Both CEOs emphasized the same shift:
“Engineering competitiveness is moving from speed and experience to strategy and orchestration.”
At Meta, senior engineers don’t write most of the logic themselves anymore.
They use an orchestration platform to assign subtasks to multiple Llama sub-models, then hand the results off to a Copilot agent for testing and deployment.
As Zuckerberg described it:
“It’s teamwork — except the team is entirely AI.”
Nadella echoed this, describing Microsoft’s “Agent Development Stack”:
- Intent input interface
- Multi-model decision layer
- Tool-access APIs
- Execution models below
“Development is no longer about what language you know — but which intelligent resources you can command.”
AI Tools Become Operational Units
To command an AI “army,” engineers must now master three new skills:
- Intent Modeling – translating human goals into executable model tasks.
- Model Orchestration – knowing which models to assign to which jobs, and how to chain or parallelize them.
- Task Supervision – monitoring quality, evaluating results, handling risk and error recovery.
Zuckerberg likened it to filmmaking:
“Each model is an actor.
Copilot is the assistant director.
APIs are the cameras.
The final product release is the movie premiere.”
It’s No Longer “What I Can Code,” But “Who I Can Command”
The hardest shift isn’t technical — it’s identity-based.
“Engineers aren’t writing code. They’re writing structure.”
— Mark Zuckerberg
Nadella added:
“Copilot isn’t your assistant. It’s your working clone.
You must learn to direct it — not compete with it.”
In this world, prompt-writing is entry level.
Orchestration is the true dividing line.
Part 3｜The Most Powerful Model Is the One That Collaborates
A key term throughout their dialogue was “orchestration.”
Zuckerberg said:
“The first generation of AI products matched one model to one task.
Now, we’re entering the era of multi-model collaboration.”
Nadella added:
“You no longer need a single supermodel that does everything — you need specialized models that talk to each other.”
From Model-Centric to Orchestration-Centric Systems
We’re shifting from:
- Gen-1 AI: one big brain for everything
- Gen-2 AI: many specialized brains working together
Each modern AI application will act like a coordination system — a chat interface in the front, a swarm of models cooperating in the back.
Microsoft is already transforming Copilot from a “code generator” into a “task orchestrator”:
- Model A interprets intent
- Model B retrieves and structures data
- Model C codes
- Model D verifies and deploys
At Meta, the “Distillation Factory” project uses modular small models for higher efficiency and control.
As Zuckerberg said:
“We’re not building an AI. We’re building an AI alliance.”
MCP, A2, LoRA — The New Grammar of Development
Nadella introduced two emerging protocols:
- MCP (Multi-Agent Coordination Protocol) — manages communication and task distribution among models.
- A2 (Agent-to-Agent Protocol) — governs how models call and confirm each other’s results.
These are like the HTTP of the AI era — invisible, but foundational.
Developers of the future won’t train one huge model; they’ll compose many models into living systems.
“AI doesn’t just talk — it holds meetings.”
— Mark Zuckerberg
Part 4｜Open Source Isn’t About Free — It’s About Control
“The value of Llama isn’t just that it’s open-source — it’s that every developer can distill and tame it into their own AI.”
— Mark Zuckerberg
He explained Meta’s internal “Distillation Factory”:
- Start with a large multimodal base model.
- Extract stable core components.
- Distill them into smaller, specialized versions for different uses and platforms.
This is not “downsizing” — it’s structural re-engineering.
Meta’s goal: empower millions to build their own assistants.
Nadella added that Azure is turning distillation into a platform capability — enabling one-click generation of customized models for enterprises.
“The power of open source isn’t sharing — it’s distribution.”
The age of massive closed models is plateauing; the structural advantage of modular, customizable AI is just beginning.
Part 5｜AI Is Rewriting Organizational Architecture
“We used to organize around people. Now we’re reorganizing around model capabilities.”
Traditional company structures existed to solve:
- Information gaps → managers
- Inconsistent decisions → bureaucracy
- Manual execution → coordination
AI is dissolving all three:
- It aggregates information
- Harmonizes decisions
- Automates execution
Zuckerberg calls this “the atomization of organizations.”
Middle layers built on “information flow” are fading away, replaced by agent-based coordination.
Inside Meta, multiple AI agents already replace cross-department meetings — exchanging tasks, tracking progress, and generating reports.
“Leaders are no longer approvers. They are designers.”
— Satya Nadella
Managers will shift from assigning people to architecting agents.
“Humans become model users.
Models become organizational participants.”
Part 6｜AI Is No Longer a Tool — It’s Your Operating System
Zuckerberg concluded:
“We used to think AI was a tool — optional, external.
Now it’s infrastructure — part of the organization itself.”
Nadella agreed:
“It’s not a plugin; it’s a new factor of production.
It changes not only how work is done, but who can do it.”
AI is becoming the second layer of infrastructure, embedded in every process.
Applications = Prompt + Orchestration + Model Structure
Documents = Conversation History + Data References + Interaction Logic
Engineers no longer build functions — they design intent structures.
The Real Divide
“The real question isn’t whether you use AI — but whether you can organize it.”
New roles are emerging:
Technical Commander, Model Orchestrator, Distillation Engineer, Agent Architect.
“It’s not about how fast you code, but how precisely you orchestrate.”
“It’s not about fancy prompts, but about building systems and managing relationships.”
The Productivity Paradox
“AI won’t raise GDP overnight because most organizations aren’t structurally ready.”
— Satya Nadella
He compared it to the history of electricity:
Factories didn’t see a productivity boom until decades later — when they rebuilt their layouts around electric power.
Likewise, AI’s real value comes when companies rebuild their processes, logic, and workflows around it.
Humans must shift from executors to system designers.
A New Beginning
When asked what gives them optimism, Nadella quoted Bob Dylan:
“You’re either busy being born, or busy dying.”
Then he added:
“The smarter choice is to stay busy rebuilding.”
“The real race isn’t about model size — it’s about who can tame AI, embed it into structure, and turn it into a new kind of productive force.”
Engineers must evolve into Model Conductors.
Managers into Process Architects.
Companies into AI-Native Organizations.
Only then will that “30% written by AI” not mark the beginning of replacement —
but the beginning of evolution:
Humans no longer as executors,
but as designers and directors of intelligent systems.
Would you like me to polish this English version into a medium-style article (native idiom, headline formatting, short paragraphs, optimized for readability and viral impact)? It can read like a professional English tech commentary piece.