Dragan Spiridonov joins the ISTA 2026 line-up, with 60 test agents and a candid bug list
He spent thirty years in IT before deciding testing needed orchestrators rather than testers. His open-source Agentic QE Fleet now ships 60 agents across 11 coding-agent platforms — and its own documentation records where the thing got it wrong.
Most people who build a testing tool describe what it does. Dragan Spiridonov’s documentation also describes the five attempts it took to work, and the point at which his agents rated a trivial issue as critical.
That is an unusual way to market software. It is a very good way to talk about autonomous systems.
Thirty years, then a rebuild
Spiridonov’s own site puts his career at more than thirty years in IT and testing, running from hardware up to what he now calls agentic quality engineering. The recent part of that arc is specific: he lists a spell as VP of Quality Engineering at Alchemy Cloud from August 2024 to September 2025, and the launch of his own boutique consultancy, Quantum QE, on 1 October 2025. He joins the ISTA 2026 line-up as its founder — his profile is at Dragan Spiridonov.
By his own account, the Alchemy Cloud work cut test generation time by 60 per cent using Robot Framework and Playwright alongside AI tooling. That figure is his, not an independently audited one, and it is worth reading as what it is: a practitioner’s report from inside one codebase.
What the fleet actually is
The clearest picture of his thinking is the code. Agentic QE Fleet is MIT-licensed and on GitHub, described in its own README as “AI-powered quality engineering agents that generate tests, find coverage gaps, detect flaky tests, and learn your codebase patterns — across 11 coding agent platforms.”
Read on 30 July 2026, the repository ships 60 agents across 13 domains plus seven test-driven-development subagents, written in TypeScript, and carried 425 stars and 77 forks. The supported platform list is long enough to be a statement in itself — Claude Code, GitHub Copilot, Cursor, Cline, OpenCode, AWS Kiro, Kilo Code, Roo Code, OpenAI Codex CLI, Windsurf and Continue.dev — as is the list of test frameworks it targets, from Jest and Vitest through pytest, JUnit and Go to Swift and Flutter.
Two more repositories sit alongside it: Sentinel, an API testing platform in Rust and Python, and nagual-qe, which he describes as self-learning quality memory.
Orchestrators, not replacements
The framing is the interesting part. On his blog, The Quality Forge, he defines the field in a single line: “Agentic Quality Engineering is the evolution from testing-as-activity to agents-as-orchestrators.” He is equally direct about what it is not — “not about replacing testers with AI”, because “agents augment expertise, they don’t replace it.”
The scaffolding for that is PACT: proactive, autonomous, collaborative, targeted. He presents it not as a maturity ladder to climb but as “a classification system for understanding and building agentic quality systems, similar to how SAE levels classify autonomous vehicles.” His consultancy site writes the same idea as PACTS, adding structured as a fifth term, and frames the goal as governance, observability and explainability of agent behaviour — “measuring confidence, not trust.”
That last phrase is the one worth arguing about over coffee in Sofia.
The part that earns attention
Anyone can publish a framework. What makes the Quality Forge write-up unusual is that it reports the failures too: agents that surfaced real security issues and memory leaks while analysing their own codebase, severity calibration that was initially “wildly inaccurate”, and five versions before the system worked. There are no effectiveness statistics in it at all.
For a conference audience that has spent two years being told autonomous testing is solved, a builder willing to publish his own miscalibration is a useful witness.
No session has been announced yet — the ISTA 2026 programme is still taking shape, and the agenda will carry the detail once it is confirmed.
