AI will – Software engineering won’t vanish under AI. Instead, the job is moving from typing code to orchestrating AI agents—while new risks like skill erosion, apprenticeship gaps, and fresh mental strain emerge.
On most teams. the change is already visible in daily work: AI agents can draft code. generate tests. scaffold services. wire up APIs. and churn out boilerplate at a speed no individual engineer can match. What’s less obvious is what happens next—when teams must decide what to build. what tradeoffs to accept. and what “done” actually means in the real world.
That’s the pivot at the heart of the debate about whether AI will kill software jobs or remake them. A senior software engineering leader who has spent two decades in big tech—including Microsoft. Snap. and Google—argues the profession’s future looks brighter than the headlines suggest. The shift may be fast enough that, within 12–18 months, fundamental changes will land across every industry that runs on software.
The claim isn’t that uncertainty disappears. “No one knows exactly where AI will take us beyond 12-18 months from now. ” the engineer says. adding that anyone insisting they do may be selling something more than insight. But the direction. the argument goes. is already clear: by 2028. the digital economy we know today will look completely different—if it doesn’t. “we’ve failed.”.
The job was never about typing
Part of the misunderstanding, the engineer says, is treating software engineering as if it were primarily about writing code. The role. he argues. has always been about solving problems—reducing complexity. minimizing the maintenance burden. and delivering something useful to the person on the other end. Code was the medium, not the purpose.
AI can handle more of the medium now. Agents can generate tests, scaffold services, and produce boilerplate. But they still can’t do the parts that determine whether a system survives contact with production: deciding what to build. understanding why it matters. and navigating the tradeoffs that keep a system resilient. secure. and aligned with real constraints.
The daily work, in this telling, moves from building line-by-line to supervising fleets of agents. Engineers will increasingly orchestrate work where agents generate business logic. analyze logs. suggest architectural changes. and write code—while human judgment focuses on mapping constraints. aligning outputs with product goals. and ensuring resilience and security.
That means the era of the specialist coder—deeply embedded in one stack, fluent in one language—is giving way to what the engineer calls the “generalist orchestrator.”
Why deep understanding still matters
If orchestration sounds like it lowers the bar, the engineer pushes back. He argues it raises it.
To manage agents effectively—especially across large codebases and complex systems—engineers need a deep understanding of underlying technology. They have to recognize what good architecture looks like. how systems fail. where performance bottlenecks emerge. and when an agent’s output is subtly wrong in ways that might not surface until production.
An orchestrator who doesn’t understand what the agents are doing isn’t orchestrating; he’s describing it as a liability. The essence of the job. he says. is shifting from writing code to developing a deep understanding of how systems work. staying curious about technical details. and focusing relentlessly on building things that meet real needs. Systems-level thinking, lived experience, and nuanced judgment applied to agent output becomes, essentially, the job description.
Three hazards the industry is not facing openly
The benefits of agentic engineering are real in the engineer’s view—but he flags three emerging hazards that need to be confronted “honestly,” even while the wider uncertainties around artificial general intelligence and societal disruption remain.
First is replenishment. If agents absorb tasks traditionally handled by junior engineers—writing simple features. fixing bugs. learning a codebase—the apprenticeship layer could shrink. Junior roles, he argues, have always been the on-ramp. If that training ground is removed. “the pipeline eventually dries up.” The industry. in his view. needs new models to replace that training: structured apprenticeships. AI-assisted internships. or systems that teach engineers to work alongside agents from day one.
Second is atrophy. The engineer says he has spoken with dozens of engineers using AI agents extensively, and many describe skill erosion. When an agent handles implementation, engineers stop building the same intuition and muscle memory. Some report it’s harder to enter a flow state—the mental mode where many breakthroughs happen. He frames this not as nostalgia but as a practical concern: engineers still need deep technical understanding to evaluate whether an agent’s output is correct. scalable. or safe. Without that, the job becomes reviewing code you can’t fully judge.
Third is exhaustion. One surprise, he says, is how mentally draining agent management can be. Engineers may get more done, but many also report greater cognitive fatigue. Constant switching between agent sessions. reviewing parallel workstreams. and maintaining coherence across semiautonomous systems creates a new kind of exhaustion—productive work that’s still intense in ways traditional engineering often wasn’t.
These aren’t theoretical problems in his telling. They’re already happening, and as adoption accelerates, they’re likely to intensify.
The economics: the cost-cutting trap
The engineer connects the productivity gains to a classic economic dynamic using the Jevons paradox. When steam engines became more efficient, people didn’t use less coal—they used more. As costs fell, new applications emerged and total demand surged. He argues the same pattern applies to AI and software engineering.
As AI makes engineers more productive, the cost of building software drops. But demand for software won’t stay flat. It will expand as companies build custom tools they previously couldn’t afford. and as backlog items that were once too expensive to tackle finally get shipped. Even problems that weren’t worth solving can become economically viable.
In the short term, he acknowledges the transition could be uneven, and for some engineers, painful. Over the long term, though, he’s confident the direction is more, not less: companies will need more engineers, not fewer.
He also argues that using AI primarily to cut costs is a strategy that tends to belong to companies already struggling. AI, he says, is accessible to everyone. That means the market will reward greater and faster innovation rather than the same level of innovation achieved more cheaply. He describes it as a tug of war: one side adds stronger players while the other side tries to save labor costs and hopes that’s enough to win. His expectation is that companies that keep expanding—while migrating entire teams to work effectively with AI—will overrun those cutting headcount.
Beyond software: a widening boundary
AI won’t just change the workflow of software engineers, the engineer argues—it expands what engineers can do. When implementation becomes cheaper and faster, engineers can think more broadly. Problems that once required months of dedicated coding can become afternoon projects.
He says the scope can move beyond software into hardware. into cross-domain systems. and into problems previously handled by other disciplines. An engineer who can orchestrate AI agents. in his view. isn’t limited to building web services or mobile apps; they can prototype physical systems. model complex processes. and solve unfamiliar problems in new industries.
It’s fun
Under all the risks and uncertainty, he makes an unusually direct point: agentic engineering is fun. Engineers aren’t leaning into the technology because they have to. he says. but because building things has always been the point. AI lets them build more, faster, and at a scale that wasn’t previously possible.
He describes the “creative leverage” of working with a capable agent as intoxicating—spinning up an idea, iterating in real time, and watching something come together in hours instead of weeks. He frames that as the reason most engineers entered the field in the first place.
Where this leads
The bottom line, he says, is that the future of work won’t be humans versus machines. It will be humans working alongside AI agents “across every role, every industry, every level of an organization.”
Engineering will look different. Daily work will change. The skills that matter will shift. But the core of the job—solving hard problems, reducing complexity, building things that work—won’t go away. If anything, he argues, it will matter more than it ever has.
The key constraint, he says, has always been capacity to solve problems, not ability to think of them. AI is about to dramatically expand that capacity. The question isn’t whether engineers are needed. It’s whether there will be enough of them.
AI software engineering AI agents labor market Jevons paradox productivity engineering skills apprenticeship orchestration digital economy