Dedy Kredo of Qodo on GPT-5's code review capabilities and enterprise AI adoption
Aug 7, 2025 with Dedy Kredo
Key Points
- Qodo has deployed GPT-5 as its default model across IDE, CLI, and Git plugins, betting that code review rather than code generation is where the model's biggest gains lie.
- GPT-5's coding improvements stem from reinforcement learning training validated against actual code correctness, with meaningful gains in enterprise languages like .NET and Java that matter for Qodo's core customer base.
- As AI agents write more code, the bottleneck shifts from generation to verification; Qodo sees code review as the next major product opportunity in developer tooling.
Summary
Dedy Kredo, CEO of Qodo (an enterprise AI coding platform), argues that GPT-5's most underappreciated upgrade is in code review, not code generation. Qodo has already made GPT-5 its default model across its IDE plugin, CLI, and Git plugin, deploying it to catch bugs before production and verify code alignment with enterprise best practices.
On what drove GPT-5's coding gains, Kredo attributes the improvement to scaling across both pre-training and reinforcement learning, with the latter being particularly significant. RL-based training with verifiable ground truth allows the model to be validated against actual code correctness at scale, producing measurable quality lifts.
Enterprise vs. Consumer Adoption
Language coverage splits by market segment. JavaScript, TypeScript, and Python dominate early adopter usage, but enterprise deployments skew heavily toward .NET and Java. Kredo says GPT-5 has meaningfully improved on those enterprise languages, which matters for Qodo's core customer base.
For enterprise agentic use cases, instruction-following precision is the critical variable. Kredo's position is that enterprise AI cannot be a general-purpose agent; it requires tightly scoped agents with defined instructions, tool sets, and permission boundaries. GPT-5's improvements in tool use and instruction fidelity are therefore more commercially relevant to enterprise buyers than surface-level benchmark gains.
The Code Review Frontier
The industry conversation has centered on autonomous code generation, but Kredo contends the challenge is shifting. As AI agents increasingly write code, the bottleneck moves to verification: ensuring generated code is tested, reviewed, and non-breaking. His view is that developers will spend less time writing code and more time reviewing it, and that this transition represents the next significant product opportunity in developer tooling. Qodo is explicitly positioning around that workflow shift.