Artificial intelligence is reshaping how software gets built. I want to share a few grounded thoughts from that shift—not hype, but how it feels to work with these tools day to day.
AI does not remove the need for human judgment in software development. It amplifies it. The better the tools become, the more important orchestration becomes: knowing what to ask, what to reject, what to refine, and when to intervene.
Among the labels currently used—vibe coding, agentic engineering, and the rest—the one that resonates most with me is agentic orchestration. The orchestra metaphor is unusually apt: many capable “players” (models, tools, checks, humans), a score (intent and constraints), and a conductor deciding when to cue whom.
The orchestra metaphor
An orchestra does not become excellent because each musician is talented in isolation. It becomes excellent because the score is understood, the timing is coordinated, the interpretation is deliberate, and the conductor knows when to guide, pause, emphasize, or correct. That is the part of the metaphor that matters most for software development with AI agents.
The analogy with coding agents maps cleanly onto this picture. On their own, these systems are extraordinarily capable across many domains: ask a difficult question, and they often answer with striking accuracy—and that baseline continues to improve rapidly. Yet to achieve meaningful results in real engineering work, we still need to learn how to harness the full strength of AI coding agents.
Instructions matter
Agentic orchestration follows the same principles as a real-world orchestra: give agents clear instructions, rehearse, guide, execute, then iterate—until you have something useful, beautiful, and efficient.
A central part of that discipline is making sure agents can see what they are doing, why they are doing it, and which objectives you expect them to meet.
In practice, this might mean asking one agent to draft a feature implementation, another to review edge cases, another to inspect accessibility concerns, and another to propose test coverage. The human role is not to accept the combined output blindly, but to decide which parts serve the product and which parts introduce noise.
“What are you looking for?” is shaped by taste, experience, and real-world usefulness—and that is how important we humans remain when we work alongside AI. Agents will offer many plausible directions; it is our job, as conductors of this digital orchestra, to guide them toward what truly matters. Answering the following questions is critical:
- Do we really need this?
- What is the goal of having this button?
- Is this layout truly helping my users achieve what they want quickly and efficiently?
The best results still come from the interaction between people who have embraced AI and the agents themselves. Without direction, you are left with a large volume of plausible but unfocused output—not necessarily what the product needs.
AI is the enabler, not the solution
AI is often presented as “the solution,” especially in a moment when the industry is racing to scale models, infrastructure, and automation. Even though this is indeed a revolution in the information technology space—as important, I would argue, as the internet in the early 1990s—AI is not the solution. AI is the enabler, not the author of the outcome. We still need people to use these tools properly to achieve meaningful results, with critical thinking and care for what actually ships.
The human role becomes more important
To see this clearly, try a simple test. Ask two different people to create something using the same AI tool—an app, a website, or a piece of writing. You will probably get two different results, because those people orchestrated the agents in different ways. Through their lenses, experience, and judgment, each final product became something its creator could reflect on as genuinely theirs.
Like any great orchestra, the beauty is not in the instruments alone, but in how they are brought together.
AI agents can generate, inspect, compare, summarize, and accelerate—but it is human judgment that gives the work direction, rhythm, and meaning.
Agentic orchestration is how we turn powerful tools into beautiful software: not by letting the machine play alone, but by conducting it with purpose, restraint, and taste.