Manos Saratsis on How Dromeas Fixes AI Code Review at Scale
Manos Saratsis — Founder, Dromeas
Manos Saratsis is the founder of Dromeas (dromeas.ai), an agentic pre-production platform built for engineering and security teams who need to move from AI-generated code to confident releases without sacrificing speed or safety. His core conviction: the tooling that helped developers write code faster never helped them trust it faster — and that gap is exactly what Dromeas is designed to fill.
The starting point for Dromeas is a shift most developers have already felt: AI coding tools removed code generation as the constraint, and immediately created a new one. As Saratsis explained to host Myles, the volume of AI-generated diffs now outpaces any team's review capacity. The result is a binary trap — teams either rubber-stamp pull requests and accept hidden risk, or they slow back down to pre-AI velocity. Neither outcome is acceptable.
The founding insight that drove Saratsis to build Dromeas was precise: the tools that accelerated writing code did nothing to accelerate trusting it. Review, verification, and confidence became the bottleneck, and existing tooling left that bottleneck untouched.
Inside Dromeas, pull-request review and trunk review run on every commit — not only at the moment a PR is opened. Catching issues earlier, before they compound, is the direct practical benefit. Rather than maintaining five disconnected tools or checklists for security, quality, compliance, tests, and documentation, Dromeas consolidates all of those dimensions into a single, unified review pass.
Critically, findings don't arrive as bare warnings. Each one comes with a repair handle — a previewed path to a fix before anything is applied to the codebase. The stated goal is that review keeps pace with how fast AI lets teams generate code, instead of becoming the thing that throttles it.
Dromeas introduces what Saratsis describes as agentic release management: release readiness is no longer a manual checklist that someone owns and occasionally forgets to run. Security findings, failing checks, compliance issues, and coverage gaps are consolidated into one verdict rather than scattered across multiple dashboards.
The aim is a release that is genuinely "complete" — not just "tests passed," but security, quality, compliance, and documentation all cleared together before anything ships.
One of the more technically distinctive aspects of Dromeas is its approach to context. Every file and function is indexed into a live call graph, so before a change is made to any function, the system can surface who calls it, what it calls, and how far the blast radius of a change might reach.
This matters because, as Saratsis argues, AI-generated changes are dangerous precisely when they are made without that broader context. A diff that is locally correct can silently break a caller three layers away. The call graph is what separates a tool that spots surface-level issues from one that understands a codebase the way a senior engineer who has been embedded in it for years would.
Dromeas does not trust any single model in isolation. Instead, it uses what Saratsis calls an "LLM council" — multiple models that must agree before something is surfaced as a real finding. The explicit purpose is to cut the false positives that erode developer trust in automated tooling over time. Teams are also not locked into one vendor's model: a bring-your-own-model capability lets organisations plug in the models they already trust or are governed to use.
On the question of how much to hand over to the system, Saratsis is clear that autonomy is a dial rather than a binary switch. One end of the dial surfaces findings for a human to review and approve. The other end auto-fixes and opens the pull request. What makes higher autonomy safe rather than reckless, he argues, is the guardrails sitting underneath it: a verification loop, human approval gates, and compliance checks.
The closing line Saratsis offered for the episode captures the philosophy behind the entire product: the goal is not to replace engineering judgment — it is to make sure that judgment is only spent where it is actually needed.
Before founding Dromeas, Saratsis built his career across engineering and product leadership. At Dynatrace, he led the Behavioural domain in Digital Experience Monitoring. Prior to that, at Netdata, he served as Director of Product and shipped the company's first cloud product from an open-source foundation. Earlier in his career he was an engineer at Vodafone Group, leading location services and the Opera Mini OEM integration programme, based in Düsseldorf.
Comments
No comments yet — be the first to share your thoughts.