The State of Software Engineering in 2026: What Changed and What Did Not

At LDX3 in September 2026, Gergely Orosz reports on conversations with AI labs, large technology companies and startups. His examples illuminate selected organizations rather than providing a representative survey of the whole industry.

Orosz observes that AI-generated code and multiple concurrent agents have become common in the organizations he visited. He cites engineers working across several tasks at once, but cautions that prominent AI companies are particularly advanced adopters.

Many companies are building their own agent harnesses. He describes tools that coordinate work, run in cloud environments and integrate with a company’s development processes; different organizations make different tradeoffs.

He gives examples of AI-assisted migrations, including Airbnb and Asana test-library transitions. Asana’s project had also benefited from years of earlier work, so short migration times should not be read without that context.

Some organizations report smaller project teams because individual engineers coordinate several agents. Orosz contrasts this with the continuing importance of teams that own software, share on-call responsibilities and cover for absences.

Planning and testing remain necessary: building the wrong thing faster is still wasteful, and generated code needs validation. He recounts a team’s experience that substantial testing effort remains alongside AI-written implementation.

Orosz says the increase in code and pull requests strains existing assumptions and infrastructure. He also observes reliability problems in products he uses; his anecdotes do not by themselves establish their causes.

He describes CPU capacity constraints and long lead times reported by companies, as well as pressure on engineering leaders. These are time-specific reports and interviews, not universal claims about every cloud region or employer.

Looking ahead, Orosz expects more cloud agents, refactoring and investment in AI maturity. He advises engineers and leaders to stay hands-on, build supporting infrastructure and use AI to learn rather than delegate away their own understanding.