Jacek Migdał is the founder and CEO of Quesma, a company building analytics and evaluation tools for coding agents. He previously spent a decade at Sumo Logic, where he helped develop its Warsaw engineering organization and tracing product. His writing explains the engineering and business decisions behind those companies.[1][2]
Start with A postmortem on our $2.5 million database gateway →
Career
Migdał joined Sumo Logic before it had revenue and stayed through its public listing. His account describes building the Warsaw office and incubating its tracing product, which follows requests through distributed software. Quesma’s current team applies related systems experience to coding agents: examining what an agent did, how much it cost, and whether its activity produced progress. Company benchmarks and research are team contributions, with authorship identified on individual publications.[1][2]
Contributions
Quesma originally built a proxy that translated Elasticsearch queries into ClickHouse SQL. In his November 2025 postmortem, Migdał explains why positive customer conversations and prominent pilots failed to create enough urgency. The technology helped database vendors close compatibility gaps, but the standalone product often ranked below customers’ immediate priorities. Quesma sold the intellectual property to Hydrolix and used the proceeds to support a pivot.[3]
Ideas
Migdał’s postmortem argues for testing customer intent with concrete actions, such as running a script, rather than treating verbal interest as demand. It also distinguishes a useful feature inside another vendor’s product from an independently urgent purchase. That makes the essay a practical account of product judgment alongside his engineering history: a technically working system can still lack a strong route to adoption.[3]
Tools, writing & talks
Writing, research & talks
Start here
A founder’s account of Quesma’s query-translation product, stalled pilots, asset sale, and pivot.
Read this before exploring Quesma’s current agent work. It explains the change in direction and gives founders a concrete distinction between an interested prospect and a customer willing to act.
Author[3]Runs one identical prompt and a six-hour budget through Opus 5.5 in Claude Code and GPT-6 Astra in Codex to build an interactive Three.js visualization of Invisible Cities, comparing design quality, runtime and cost.
Quesma article by Piotr Migdał; associated through Jacek’s companyOriginal sourceAn empirical benchmark of Qwen3.8 27B GGUF quantizations on GPQA Diamond, IFBench, and Terminal-Bench 2.1 finds 4-bit (17GB, fits a 24GB card) matches full-precision performance, while 1-bit quantization collapses toward random accuracy.
Quesma article by Piotr Migdał; associated through Jacek’s companyOriginal source