How BASF delivered a new platform with 70% less effort using a team of Claude agents

BASF's Dlite portal serves the company's cosmetics and healthcare sector. Adding anything new to it means working within established design patterns, strict security rules and tight access controls.

When BASF needed a new event management platform built into Dlite, the challenge went beyond writing code. The team had to carry business intent from stakeholder meetings through to tested, documented, production-ready APIs, without losing context at every hand-off.

Traditional delivery handles this in a straight line: requirements, then design, then build, then test, then document. Each stage is a hand-off, and each hand-off is a chance for the original intent to get diluted.

Tenth Revolution Group took a different approach.

One project, six specialist agents

TRG's engineers designed a multi-agent workflow on Anthropic's Claude that covered the full development lifecycle. Each stage was handled by an agent with a defined role:

• Architectural analyst: turned raw meeting transcripts and notes into the system architecture, Azure cloud requirements, the logical data model and a full backlog of user stories.
• Infrastructure and data engineer: converted those stories into database schemas, migration scripts and scripted Azure Storage provisioning.
• Senior backend developer: built the core APIs under system prompts that enforced Dlite's clean code standards, design patterns and access control rules.
• QA and test automation: generated unit, integration and regression test suites against the finished code.
• Technical writer: produced readable use-case documentation directly from the codebase and the architecture decisions.
• Data operations developer: built a CSV import tool so administrators could publish events before the admin dashboard was finished.

Each job used the right model for it. The most capable Claude model handled architecture and reasoning, a mid-tier model handled core development, and a fast, lightweight model covered testing and documentation. Prompt caching kept Dlite's security guidelines and design rules in context at every step, which cut cost and latency without the agents losing sight of the standards they had to meet.

TRG owned prompt engineering, orchestration and technical delivery. BASF provided domain knowledge, infrastructure access and user acceptance testing sign-off.

What it meant for BASF

• 70% less delivery effort. Across architecture, development, testing and documentation, the agent-driven approach cut total effort by around 70% compared with traditional delivery. The biggest gains came in testing and documentation, where effort fell by nearly 90%.
• Live in under ten weeks. The platform went from kickoff in May to production in July 2026, on schedule.
• Complex work in half the time. The most complex features, work that would typically fill a full sprint, were delivered 50–60% faster.
• Events published from day one. Administrators could load and publish events at go-live without waiting for the admin dashboard.
• Quality built in, not bolted on. Near-100% automated test coverage meant the platform was certified stable before launch, with minimal manual testing effort.
• A model that lasts. BASF continues to use the same AI-driven workflow for testing and documentation on future Dlite releases, so the gains carry into every iteration.

Why it matters

AI-assisted development is usually talked about as a coding productivity tool. The Dlite project shows something bigger. When agents are orchestrated across the whole lifecycle, the context that normally leaks out at each hand-off stays intact, from the first stakeholder meeting to the final line of documentation. The result is software that ships faster, is better tested and is easier to maintain.