EXANTE's Codey completes 75% of 250 tasks; humans keep release control
For the CIO of a regional bank piloting coding agents
EXANTE, an investment firm, said 250 tasks have passed through Codey, its in-house multi-agent AI developer, since launch and 75% were completed successfully, according to a FinTech Global report dated Oct. 16.1
According to EXANTE, its head of web applications, Alina, led the build as part of a wider AI transformation programme, after earlier tools such as AI code review sped up single stages but stopped short of full-cycle development.1 Codey runs on Claude Agent SDK, Codex and OpenCode and coordinates six agents: a researcher, planner, developer, reviewer, QA tester and verifier, with reviewer-flagged work sent back for revision.1
Codey now handles 7% of all work in the teams that use it and typically moves from the start of coding to a deployed feature environment in around 30 minutes, according to the report.1
Codey is limited to three revision cycles before a task is marked "AI failed", and humans keep responsibility for product acceptance and release decisions.1
EXANTE says most failures stem from missing context, changes spanning multiple interdependent repositories, or first tasks in an unfamiliar project or stack.1
Tasks arrive through sprint planning or Slack, where Codey clarifies requirements, raises a Jira ticket and returns a merge request and preview link; EXANTE uses it on internal CRM and back-office systems and on its client portal, web trading terminal and desktop application.1
For banks weighing AI coding agents, the cap on revision rounds and human sign-off show where control sits, and EXANTE reports a temporary shift of workload toward review. Review capacity, requirement quality and unfamiliar or interlinked codebases, not model output, may set delivery speed.
EXANTE plans automated backlog analysis, wider departmental coverage, richer context sources and feature-based delivery, aiming to hand Codey the full cycle from task definition to production; it gave no dates.1
Our take. One in four Codey tasks failed, and every release still waits on a human signature. That is the honest ceiling for agent autonomy in your change-management policy, whatever the demo shows.
Action. Change management: Require named human approval and a capped revision count on every agent-authored merge request before production release.
What our records say
Claude's posture, per its programme summary, is selling banks seats and API access directly and through cloud marketplaces, and in 2026 adding embedded engineering with Goldman Sachs and FIS, restricted Mythos access for a few large banks, and a services firm co-founded with Goldman Sachs.1
From our records
- Claude (Anthropic)vendor · banks using it (8) Goldman, JPMorgan, BankUnited · alternatives Microsoft Copilot, ChatGPT Enterprise, Gemini Enterprise
Sources
- ReportEXANTE’s Codey hits 75% success, but humans keep control FinTech Global