|
No invasive meeting bots (Sponsor)
Ever had a meeting where a random bot joins the call and, suddenly, everyone’s distracted?
Granola works differently.
There are no meeting bots. Nothing joins your call.
Granola transcribes directly from your device’s audio (on your computer or your phone). It works with any meeting tool: Zoom, Google Meet, Microsoft team … and even for in-person conversations.
You stay focused and jot down notes like you normally would.
Granola quietly transcribes and enhances the important bits in the background.And if you want to be extra thoughtful or compliant, you can always send a quick consent email beforehand - automatically disclosing your use of Granola.
Try Granola on your next meeting and see how much easier it is to stay present.
The tech industry has solved code generation but created a massive verification crisis.
This time, let’s dive into Sonar’s 2026 State of Code Developer Survey, conducted across 1,149 developers globally in October 2025. The polls reveals that enterprises are now writing code 10 times faster while simultaneously choking on the review burden—and most organizations haven’t adapted.
Speed Without Confidence
The data is stark and unforgiving. AI now accounts for 42% of all committed code, a figure developers predict will reach 65% by 2027—up from just 6% in 2023.
Among developers who have tried AI tools, 72% now use them daily. This adoption is near-total: AI is deployed across prototypes (88%), internal production (83%), customer-facing applications (73%), and even mission-critical services (58%).
Yet here is the contradiction that defines 2026 software engineering: 96% of developers do not fully trust that AI-generated code is functionally correct—but only 48% always verify it before committing. This gap between distrust and action is what AWS CTO Werner Vogels has termed “verification debt,” the accumulated burden of unreviewed code that will eventually demand repayment.
The verification step is not trivial. Nearly all developers (95%) spend effort reviewing and testing AI output; a majority (59%) describe that effort as “moderate” or “substantial.”
Most damaging to the productivity narrative: 38% of developers say reviewing AI-generated code requires more effort than reviewing code written by human colleagues—a finding that inverts the promised efficiency gain.
Toil Didn’t Disappear; It Migrated
One of the most revealing findings from the Sonar data contradicts the cheerful narrative about AI’s impact on developer well-being. Developers report spending approximately 24% of their work week on toil—repetitive, frustrating tasks that hinder productivity—regardless of how frequently they use AI.
The important nuance: the nature of that toil has shifted. Developers who use AI rarely are more likely to cite “debugging legacy or poorly documented code” as their primary frustration; developers who use AI multiple times per day report their biggest sources of toil as “managing technical debt” and “correcting or rewriting AI-generated code.”
This shift matters because it reveals a truth the industry hasn’t fully absorbed: AI didn’t eliminate friction in software development; it relocated it downstream to the verification stage. The time developers save drafting code is reinvested directly into ensuring that AI-generated code meets production standards.
A developer coding faster is simply creating more work for the review stage—work that is fundamentally harder because AI-generated code, by the admission of 61% of developers, “looks correct but isn’t reliable.”
Junior Developers Bear the Heaviest Burden
The Sonar data exposes a generational divide in how AI impacts developers. Junior developers (≤10 years of experience) report the highest productivity gains—40%—and are simultaneously the most burdened by review complexity.
While this cohort is reporting 40% productivity improvements, they face a paradox: 66% say that AI-generated code looks correct but is unreliable, and 40% report that reviewing AI code requires more effort than reviewing human-written code. These developers are experiencing speed on the generation side while suffering complexity on the verification side.
By contrast, senior developers (≥20 years) report 32% productivity gains—more modest—but face less friction during review. The pattern suggests that experienced developers are more selective in their AI usage, deploying it for specific high-impact tasks (documentation, test generation, code explanation) rather than as an always-on assistant.
This generational divide has strategic implications: junior developers are the fastest adopters and highest productivity gainers, but they’re also the least equipped to catch the subtle bugs that characterize AI code failure.
Organizations that fail to invest in verification guardrails will inadvertently make their most junior engineers the ones responsible for catching the most difficult problems.
The Double-Edged Sword
The relationship between AI-generated code and technical debt is complex and dangerous. Developers report mixed outcomes. On one hand, 93% cite positive impacts from AI: 57% see improved documentation, 53% see better test coverage. These are genuine improvements. But simultaneously, 88% report negative impacts from AI use: 53% say it generates code that looks correct but isn’t reliable; 40% report it creates unnecessary and duplicative code.
The danger is that these problems compound silently. When developers can generate code in minutes rather than hours, they’re far more likely to accept a suboptimal solution and move forward. Small architectural inconsistencies, redundant implementations, and subtle reliability issues accumulate in the codebase faster than human review can detect or address them. This is not a theoretical risk—it’s already visible in the data. SMB developers report spending 65% more time correcting AI code compared to enterprise developers, suggesting that without governance infrastructure, speed rapidly converts to rework burden.
Shadow AI and Governance Failure
One of the most alarming findings in the Sonar report concerns developer tool adoption patterns. The average development team now juggles four different AI coding tools, and 35% of developers are accessing these tools through personal accounts rather than work-sanctioned ones. This isn’t accidental—it reflects the reality that developers are urgently experimenting with tools and moving faster than official governance processes can authorize.
The security implications are severe. When a developer uses ChatGPT via their personal account to generate code, their company’s security team has no visibility into what proprietary data or sensitive logic may have been exposed to public models. With 52% of ChatGPT users and 63% of Perplexity users accessing those tools through personal accounts, this represents a massive blind spot for compliance and security teams.
This fragmentation also undermines standardization efforts. If different teams are using different AI tools with different implicit biases, code review standards become inconsistent. The “bring your own AI” culture that’s emerged is pragmatically efficient but systemically risky—exactly the opposite of what enterprise governance should allow.