Artificial intelligence (AI) in the software development lifecycle (SDLC) isn’t a marketing slogan for us – it’s something we constantly measure. We run AI-SDLC at two levels: a baseline level, mandatory on every project, and a full level wherever the project and the client allow it.
The baseline level is writing specifications before code (spec-driven development, SDD) the only change to the old process, with nothing else automated: we still write pull requests and do review by hand. Our internal analysis shows that this change alone cuts ticket completion time by about 30% against a baseline calibrated from each developer’s own history a speed-up of roughly 1.2 to 1.5 times. We confirmed the result across several variants of the analysis, not from a single measurement.
Wherever the project and the client give us full latitude, we run full AI-SDLC: agentic orchestration during implementation, review carried out independently by a different model than the one that wrote the code, and tests generated and run by AI (plus a traditional check at the end). In this setup we’re currently running a migration that would take many months in a traditional, non-AI approach. The tickets completed so far add up to 66 hours and 45 minutes of estimated time against 25 hours of actual work 2.7 times faster, a 62% reduction in time.
We add Fuzberus to this picture – our in-house system for AI-based pull request checking. It has been running in production for many months; wherever the process requires a human PR, Fuzberus takes over the duties of the first gate, to take the load off the developer.
