AI-Powered Software Development Delivers Measurable Results
In a June 12, 2026 Forbes Technology Council column, Fabio Caversan, global CTO at IT services firm Stefanini, argues that AI-assisted coding has moved from simple autocomplete to structured, spec-driven execution that can compress software delivery timelines. As a self-reported example, he says a legacy Java/PowerBuilder modernization that took 8-12 weeks and roughly 1,280-1,920 engineering hours with a four-person team in 2022 has since shrunk to about two weeks and 120-200 hours using AI-driven multi-agent orchestration. The piece is Caversan's own opinion and vendor case study, not independently verified reporting, and its own caveat is that coordination, governance, and shifting priorities, not the AI tooling, are now the harder bottleneck to fix.
The concrete number here is the one worth remembering: a legacy-modernization job that reportedly took 1,280-1,920 engineering hours in 2022 is described as needing only 120-200 hours today with AI multi-agent orchestration, according to the vendor executive who wrote this piece. That is a useful directional data point for engineering leaders sizing AI-assisted modernization work, but it is a single self-reported case from an IT-services executive whose firm sells these services, so it deserves normal vendor-benchmark skepticism rather than being treated as an industry-wide average.
What happened
Fabio Caversan, global CTO at IT-services firm Stefanini Group, writes in a June 12, 2026 Forbes Technology Council column that AI-assisted software development has moved from basic autocomplete into structured execution models built around spec-driven development: AI agents work from explicit requirements, architectural constraints, and validation checkpoints rather than ad hoc prompts. He argues this improves traceability, repeatability, and predictability, and, in well-scoped work such as software modernization, can meaningfully cut delivery time and cost.
Technical context
As his illustrative example, Caversan describes a Java 7/PowerBuilder legacy-modernization project. In 2022, engineering-only effort (excluding onboarding, governance, security, and infrastructure work) ran 8-12 weeks with a four-person team, about 1,280-1,920 hours. By 2023-2024, with AI embedded in the workflow, he says the same scope dropped to roughly 4-6 weeks and 480-720 hours. With structured multi-agent orchestration, separate AI agents handling architecture interpretation, code generation, testing, and documentation in parallel under shared project context, he says it now takes about two weeks and 120-200 hours.
For practitioners
Caversan's framing is a useful checklist for teams: pair AI execution with explicit specs, architectural guardrails, CI/CD integration, and validation loops rather than open-ended prompting. He is also candid about the limits - AI does not fix multi-team coordination problems, cross-functional dependencies, or unclear ownership, and does not automatically create alignment. That caveat matters more than the productivity numbers for most enterprise teams.
What to watch
This is a single-author opinion piece from an executive at a company, Stefanini, that sells AI-modernization services, published as contributor Council Post content rather than Forbes newsroom reporting. Treat the hour and week figures as a vendor's own illustrative case study rather than an independently benchmarked result, and watch for third-party studies that test whether these compression ratios hold across other codebases and teams.
Key Points
- 1Forbes Council contributor Fabio Caversan, Stefanini's global CTO, argues AI coding has shifted from autocomplete to spec-driven, multi-agent execution models.
- 2His illustrative case study claims a legacy modernization project shrank from 1,280-1,920 engineering hours in 2022 to 120-200 hours with AI orchestration.
- 3The piece is vendor-authored opinion content, not independently verified reporting, and its own caveat is that coordination and governance remain the harder bottleneck.
Scoring Rationale
Pulled from 5.8 to 4.8: this is a single-author vendor opinion column (Forbes Council contributor content, not Forbes reporting) built around one self-reported, unverifiable case study from an executive whose firm sells the services described. The underlying practitioner pattern (spec-driven, multi-agent AI development) is real and relevant, but there is no independent corroboration of the specific hour/week figures, warranting a solid-but-modest score rather than notable.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
