Thomas Ryd is CEO & cofounder of Northern.tech, a device lifecycle management leader with a mission to secure the world’s connected devices.
getty
The headline story of the past two years is that AI is coming for human jobs. Take software engineering. Marc Benioff announced Salesforce was considering hiring zero new engineers in the future, citing AI productivity gains. Entry-level developer hiring has dropped 20% since 2022.
Read only those numbers, and the conclusion seems obvious: AI can code, so there is less need for people to do the same.
A closer look at the same market data tells a different story. On Indeed, job postings for software engineers are up 11% annually, outpacing overall postings. Bank of America’s employer survey found that companies are expanding software budgets and increasing engineering headcounts. IBM is “tripling entry-level hiring in the United States, including software developers.” And across the industry, the roles growing fastest are not the ones furthest from AI; they are the ones closest to accumulated domain knowledge—where humans and technology intersect.
The question worth asking is not whether AI can write code. It clearly can. The question is whether AI can replicate what mature software companies actually sell: not code, but solutions.
Can AI Build Enterprise-Grade Software?
An AI model can credibly generate code that appears to be an enterprise platform. It cannot, however, recreate the decades of intelligence that went into it—partnerships, strategies, security, compliance, behavioral patterns and usage. Code is replicable. Context is not.
Enterprise-grade products are built on decisions made by teams with a full-time focus on a single problem domain throughout their entire careers. This expertise does not exist in AI training data. Some of it was never documented, some was never digitized and some is proprietary and lies behind NDAs. It lives in design reviews, support escalations, conversations and the quiet judgment of engineers who have seen the same failure mode countless different ways.
The True Story Behind Hiring Data
Companies are not eliminating technical roles. They are reshaping them around the activities AI cannot do well: supplementing human expertise rather than replacing it.
Technical Expertise To Inform AI Decisions
McKinsey’s research shows AI tools delivering 20% to 45% productivity gains on routine coding tasks. However, the same tools struggle with system design, architectural decisions and understanding business context. Predictably, organizations are cutting where AI handles output and hiring where AI needs human guidance. Atlassian’s restructuring offers a clean example: the company reduced 1,600 positions while searching for new hires in AI engineering, ML operations and AI safety.
The pattern repeats elsewhere. Entry-level and generalist developer roles have contracted. Yet, senior engineer openings in AI, cloud infrastructure, cybersecurity and legacy modernization remain persistently high throughout the layoff cycle. Demand for AI governance skills is up 150%. AI ethics roles, which did not exist as a category three years ago, are up 125%.
If AI generates the foundational code that junior engineers once handled, how will the next generation develop the instinct to troubleshoot and recognize when a model is producing bloated or subtly flawed output? Today’s senior engineers built their judgment on years of writing and debugging the low-level work now handled by AI. Removing the junior layer may create a talent pipeline problem that only surfaces years down the line.
Relational Expertise Defines Enterprise Success
Notably, customer-facing and relational functions are not seeing the same pressure. Across leading software producers, the developer role has evolved from routine coding to working directly with customers and scoping features that AI can produce. The valuable humans in the loop are the ones translating between the messy reality of a customer’s problem and the structured input an AI tool can act on.
AI handles what it was trained on. Humans remain essential for relational cues in a sales conversation, the undocumented workaround a long-time user relies on, integration requirements, edge cases and unwritten best practices. These are the hardest problems enterprise software solves, and they require humans with real-world expertise, not AI models trained on generalized data.
AI Strengthens The Value Of Enterprise-Grade Software
While AI accelerates code production, it struggles to replicate the complex, domain-specific systems that underpin regulated industries, connected devices and mission-critical infrastructure. These solutions are not collections of basic functions. They are the product of the lessons learned from thousands of real-world deployments. No model trained on public data can replicate that.
For organizations evaluating software solutions, AI-assisted or not, the question remains: Is this capability a core differentiator for our business? For product-differentiating capabilities, AI tools can meaningfully extend an internal team. For non-differentiating solutions, like infrastructure systems, the value of purpose-built, enterprise-grade software has grown, as the gap between what AI can generate and what mature software solutions deliver remains significant.
Redefining The Value Of Artificial And Human Intelligence
AI implementation remains in full swing. However, it is not replacing the need for enterprise software teams. The well-documented, pattern-matchable work is being absorbed into AI tools at a steady pace. The work that remains, and continues to demand investment, includes system architecture, domain expertise, the relational elements and the governance of AI systems themselves. Strategically, OEMs cannot simply replace engineers with AI tools. The better approach is to redirect human investment toward the deep contextual understanding, architectural judgment and day-to-day expertise that a model cannot match.
The AI-driven restructuring underway is not a signal that software companies are becoming obsolete but that their composition is changing. AI changes how software gets built; it does not change why buyers choose one solution over another. That choice is still made on trust and the confidence that the team behind the product is already anticipating the challenge the buyer has not yet thought of.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
Read More
Thomas Ryd