AI has been the loudest character in the jobs conversation for more than a year now. In headlines and hallway debates alike, the story is familiar: machines get smarter, software gets cheaper, and the people who used to build it are suddenly “redundant.” Engineering—often treated as the most directly exposed occupation because it sits closest to code, automation, and productivity tools—has been at the center of that anxiety.
But new hiring data is complicating the narrative in a way that’s hard to ignore. According to SignalFire’s analysis of startup hiring patterns, engineers as a share of total new hires have actually increased, even as AI continues to drive layoffs and hiring freezes in some companies and functions. The implication isn’t that job losses aren’t happening. It’s that the relationship between AI and engineering employment may be less linear than the popular storyline suggests.
This is the kind of mismatch that tends to emerge when people interpret technology change through a single lens. If you only look at reductions—headcount cuts, role eliminations, “we’re consolidating teams”—you can conclude that AI is replacing labor across the board. But if you also look at where companies are still spending money, what they’re still building, and which roles they’re still adding, a different picture appears: engineering demand may be shifting rather than disappearing.
To understand why, it helps to separate three ideas that often get blended together: automation, augmentation, and reallocation.
Automation is the part everyone talks about. AI systems can write code, generate drafts, summarize documentation, and automate certain workflows. That can reduce the number of people needed for specific tasks. In some organizations, it does lead to layoffs or slower hiring.
Augmentation is the part that’s easier to underestimate. Even when AI reduces the time required for a task, it doesn’t necessarily reduce the number of tasks. Instead, it can increase throughput. Teams can ship more features, iterate faster, and expand into new product surfaces. When that happens, the bottleneck moves. The work doesn’t vanish; it changes shape. More engineering effort goes into integration, quality, security, observability, and the glue that turns prototypes into reliable systems.
Reallocation is the third piece. Companies don’t just decide whether they need “engineering.” They decide which kinds of engineering matter most right now. AI can compress some categories of work while expanding others. A company might hire fewer people for one type of development and hire more for another—especially around data pipelines, model evaluation, infrastructure, platform tooling, and customer-facing reliability.
SignalFire’s finding—engineers increasing as a share of new hires—suggests that, at least in the startup ecosystem SignalFire tracks, the net effect is not a simple replacement of engineers with AI. Instead, AI appears to be pulling more hiring weight toward engineering roles relative to other functions.
That doesn’t mean every engineer is safe, or that every team is growing. It means the composition of hiring is moving in a direction that contradicts the “AI kills engineering jobs” framing. And that contradiction matters, because it points to a deeper truth about how technology adoption actually plays out inside companies: the first wave of AI adoption often creates new engineering work before it eliminates old work.
The “first wave” effect: building the bridge before removing the scaffolding
When AI tools become widely available, many organizations start with experimentation. They test copilots, evaluate model APIs, run pilots, and try to identify where AI can reduce costs or improve speed. Those pilots rarely succeed without significant engineering involvement. Even if the model itself is external, the surrounding system is not. Someone has to connect it to products, manage latency, handle failures, enforce permissions, design prompts and workflows, and ensure outputs meet quality standards.
In other words, the early stage of AI adoption is often an engineering project, not a labor replacement event. Companies need engineers to operationalize AI. They need them to turn “cool demo” into “works reliably for real users.”
Once that bridge is built, the next stage is optimization. At that point, some roles may shrink. But optimization typically targets specific tasks, not entire job families. It’s common to see productivity gains translate into fewer hours per feature rather than fewer engineers overall—at least until the organization reaches a saturation point.
SignalFire’s data aligns with this pattern. If engineering share of new hires is rising, it suggests companies are still in the phase where AI adoption requires additional engineering capacity. They may be hiring engineers not because they want to “employ more engineers,” but because they need the technical muscle to integrate AI into their core operations.
Why “engineers” might be increasing even if “AI roles” are changing
Another reason the headline narrative can mislead is that it treats “AI” as a single job category. In reality, AI affects many job categories simultaneously. Some companies create new roles—model trainers, ML engineers, AI product managers. Others don’t. Some outsource parts of the stack. Others build in-house.
But even when companies don’t hire more “AI specialists,” they still need engineers to make AI usable. That includes:
Engineering for data readiness: cleaning, labeling, governance, and pipeline reliability.
Engineering for evaluation: measuring model performance, hallucination rates, safety constraints, and drift.
Engineering for integration: connecting AI outputs to existing systems, databases, and user interfaces.
Engineering for security and compliance: controlling access, logging, and audit trails.
Engineering for reliability: monitoring, incident response, and fallback strategies.
These are not always labeled as “AI jobs,” but they are engineering jobs. So if you track engineering hiring broadly, you can see growth even if the “AI job creation” story is uneven.
There’s also a subtle shift in what counts as “engineering” in hiring data. Many companies historically hired engineers for product development and infrastructure. As AI becomes a product layer, those same engineers increasingly work on AI-adjacent systems. That can raise the share of engineering hires even if the total number of hires fluctuates.
The layoff narrative vs. the hiring reality: two different clocks
Layoffs tend to be visible quickly. A company announces restructuring, headcount reductions, or a hiring freeze. Those events are dramatic and easy to report. Hiring is slower and more distributed. It happens through multiple channels, over longer periods, and often without the same level of public attention.
So it’s possible for both things to be true at once: layoffs occur in some areas while hiring continues elsewhere. The question is whether hiring is shrinking overall, and whether engineering is being cut disproportionately.
SignalFire’s data suggests that, at least in the aggregate it measures, engineering is not being cut in a way that reduces its share of new hires. Instead, engineering’s relative presence is increasing. That doesn’t eliminate the possibility of layoffs; it suggests that the labor market is reallocating rather than contracting uniformly.
This is consistent with how startups behave. Startups often run lean. When they cut, they cut selectively. They might remove roles tied to non-core initiatives, reduce duplicated functions, or pause hiring in areas where revenue impact is uncertain. But if AI is becoming core to product differentiation, then engineering becomes central again—especially for teams that can deliver measurable improvements.
A unique take: AI may be increasing the “engineering surface area”
One of the most interesting implications of the data is that AI might be expanding the surface area of engineering work. That sounds counterintuitive—if AI automates tasks, shouldn’t the surface area shrink? But consider what happens when AI becomes embedded in products.
Every AI feature introduces new failure modes and new requirements. You don’t just need to generate text or predictions. You need to handle edge cases, prevent unsafe outputs, manage user expectations, and ensure the system behaves consistently across contexts. You need to build guardrails, evaluation harnesses, and monitoring dashboards. You need to design user flows that degrade gracefully when the model is uncertain.
That means AI doesn’t simply replace a chunk of coding. It adds layers of complexity around the model. The engineering work shifts from writing everything from scratch to building systems that can safely and reliably use AI.
In that world, the “unit of engineering value” changes. A small team can do more, but the team needs to be more specialized in system design, testing, and operational excellence. That can increase demand for engineers who can bridge product and infrastructure.
So while AI may reduce the need for certain types of implementation work, it can increase the need for engineers who can orchestrate complex systems. That would naturally raise engineering’s share of new hires.
What this could mean for workers: fewer “generic” roles, more “systems” roles
If engineering hiring share is rising, it doesn’t automatically mean the experience of engineers will be uniform. The composition of engineering work may be changing. Companies may be hiring engineers who can:
Work with AI APIs and model providers.
Design evaluation and quality measurement.
Build data pipelines and governance frameworks.
Implement safety, privacy, and compliance controls.
Create robust monitoring and feedback loops.
Integrate AI into production systems with strong reliability guarantees.
Meanwhile, roles that were previously more focused on repetitive coding tasks may face more pressure. That doesn’t mean those roles disappear overnight. It means the market may reward different skills.
For job seekers, the practical takeaway is that “engineering” is not a monolith. The market may be hiring engineers, but it may be hiring them for different reasons. The best strategy may be to align with the engineering work that AI adoption creates: integration, evaluation, reliability, and responsible deployment.
For employers, the takeaway is that AI adoption is not a cost-cutting switch by default. It’s a transformation project. Transformation projects require engineering capacity. Even if AI reduces marginal costs later, the transition period can increase hiring needs.
The SignalFire angle: why startups are a key lens
SignalFire’s data focuses on startups, which matters because startups often move faster than large enterprises. They adopt new tools quickly, and they feel competitive pressure intensely. If AI becomes a differentiator—better customer support, faster content generation