Taller Technologies’ Chiron platform cut three modernization programs from 36 to 9.3 summed project-weeks, with gains coming only once agents were orchestrated
Two Taller Technologies executives report a retrospective field study of Chiron, their platform that coordinates humans and AI agents through analysis, planning, implementation and validation. It covers three real modernization programs: a COBOL banking system moved to Python, Next.js and PostgreSQL (about 30k lines), a large accounting modernization (about 400k lines), and a .NET 3 to .NET 8, Angular to React mortgage application (about 30k lines). Across a traditional baseline and four platform versions, summed duration fell from 36.0 to 9.3 project-weeks, issues reaching validation fell from 8.03 to 2.09 per 100 tasks, and first-release coverage rose from 77.0% to 90.5%. The early tool-centric versions were faster but worse on quality; gains came with acceptance-criteria validation, repository-native review and hybrid human-agent execution.
The numbers (as reported)
| Summed project duration | 36.0 weeks (traditional) to 9.3 weeks (V4), 3.87x faster |
|---|---|
| Validation-stage issue load | 8.03 to 2.09 issues per 100 tasks |
| First-release coverage | 77.0% to 90.5% (task-weighted) |
| Modeled effort | 1,080 to 232.5 person-days |
| Review-stage containment (V4) | About 51.4% of issues caught before validation |
| Programs | 3: about 30k, 400k and 30k lines |
Facts
- Company / subject
- Taller Technologies
- Tools
- Unstated
- In-house tools
- Chiron
- Codebase
- brownfield
- Work type
- Upgrade / modernisation
- Stack
- COBOL, Python, Next.js, PostgreSQL, .NET, Angular, React, GraphQL
- Published
- 2026-03-17
- Last checked
- 2026-11-13
What the source claims Practitioner (own blog/paper)
Claimed by the builder, writing about their own work. Not checked independently.
- Summed delivery time cut from 36.0 to 9.3 project-weeks 3.87x speedup
- Issues reaching validation fell from 8.03 to 2.09 per 100 tasks 74.0% reduction
- First-release coverage rose from 77.0% to 90.5% +13.4 pp
Caveats
Both authors are executives at the company that built Chiron and disclose the conflict of interest. The benchmark was assembled retrospectively from engineering records and practitioner recall, in one organization, with no independent audit. Person-day figures are modeled from staffing assumptions, not logged hours. Models and coding tools are not named.
More stories by tool: Unstated
Related topics: Upgrades and modernisationBrownfield codebases
Source
arXiv (Maximiliano Armesto and Christophe Kolb, Taller Technologies)