ai/agents ai/enterprise aiadoption casestudies engineering wso2 conference
Core idea
Scaling AI across an organization requires a shared operating model, not only access to AI tools. Amila Mahaarachchi describes WSO2’s move from fragmented coding assistants to a common platform with product-specific skills, agent-friendly repositories, repeatable workflows, human review, and internal champions. WSO2 reports productivity and consistency gains, while describing its broader agentic-enterprise program as still in progress.
Video
1. Why WSO2 consolidated its AI tools
Before 2026, engineers could choose tools such as Claude Code, GitHub Copilot, ChatGPT Codex, or Cursor. Mahaarachchi says this created uneven results: outcomes depended on both the tool and the engineer’s skill. Teams also risked duplicating work instead of sharing solutions.
WSO2 chose a common platform after reviewing staff feedback and doing its own research. The goal was not only to standardize the tool. WSO2 also wanted a shared foundation for reusable knowledge and collaboration.
2. Build context and workflow into the system
WSO2’s engineering organization supports more than 40 product versions, including established products spread across many repositories. A coding assistant could miss important context if it could not navigate across repositories or follow product-specific standards.
The team addressed this with:
- Product and company skills: Shared instructions that give the assistant relevant WSO2 and product context.
- Agent-friendly repositories: Guidance to help assistants navigate mature, multi-repository products and apply design and security rules.
- Codified maintenance steps: Product-specific skills for reproducing a bug, planning a fix, assessing risk, and verifying the change.
The stated purpose is to make maintenance work more consistent and less dependent on each engineer’s individual AI expertise.
3. Automate bounded work, keep human ownership
Documentation fixes
WSO2 connected documentation issues to a GitHub Actions workflow that passes a proposed fix to a coding assistant. The assistant can make a change, verify it, and open a pull request. A person still reviews and approves the change.
Mahaarachchi reports that a documentation fix which could take about 30 minutes now takes about five minutes of human effort. He also reports more than 300 documentation issues resolved over eight to nine months. These are speaker-reported results, not independently validated measurements.
Product code and security
The engineering team uses assistants to help investigate bugs and scan code or security reports. Repository rules provide guidance, but engineers remain responsible for submitted pull requests. The process includes manual review and an AI review tool.
Mahaarachchi distinguishes internal applications from customer products: WSO2 is more comfortable with rapid, AI-heavy development for internal applications, but he says the company is not yet ready to use “vibe coding” for product development.
4. Other reported applications and targets
| Use case | Reported status |
|---|---|
| Customer support agent, Novara | Target: reduce submitted support tickets by 30%. Reported reduction at the time of the talk: 5%. |
| Customer reliability engineering assistant | Intended to help troubleshoot issues and find root causes. The desired mean time to resolve target is not yet achieved; the transcript’s numeric target is unclear. |
| SRE incident alerting | Combines multiple alerts into one incident to reduce alert overload. |
| Employee assistant | Gives employees a chat interface to internal policy documents and applications. |
| Marketing customer profiling | Used to identify ideal customer profiles for targeted marketing. |
Targets are not results
WSO2’s 30% support-ticket reduction is a target, not achieved performance. The reported result is 5%. The speaker also says some initiatives remain early and that the company is still collecting impact data.
5. Adoption, governance, and cost
WSO2 formed a virtual team with engineers from different product groups. It selected a practical priority, set weekly or biweekly milestones, released changes quickly, gathered user feedback, and used team members as adoption champions. One engineer who initially disliked the bug-fixing assistant later reported using it for 20% of updates.
The talk reports about 40 updates per month created with agent assistance. Mahaarachchi also says WSO2 resolved 300-plus documentation issues. The company uses a team subscription and says this avoids unexpected spending. It reports an 85% cost saving compared with the enterprise plan, while using WSO2’s AI gateway for additional insights and guardrails. Gateway use was partial at the time of the talk.
These figures are company-reported. The speaker says broader impact measurement is still underway.
6. What remains on the roadmap
WSO2 describes itself as partway through its agentic-enterprise journey. Planned work includes monitoring agents through its agent platform, expanding use of the AI gateway, connecting agents to each other, and exploring spec-driven development on new projects. Mahaarachchi presents these as future plans, not completed capabilities.
His recommendations are to define a clear objective, break delivery into small milestones, choose technology based on current needs, and improve through fast releases and feedback.
Key takeaways
- Standardize useful context, not only the assistant. Shared skills and repository guidance can reduce variation between users.
- Start with a concrete workflow. WSO2 focused engineering work on maintenance tasks and documentation fixes.
- Keep people accountable. AI-assisted changes still go through human review, and engineers own submitted code.
- Use champions and feedback. Short delivery cycles helped teams refine tools and support adoption.
- Separate targets from measured outcomes. WSO2’s support-ticket target is 30%; the reported reduction was 5% at the time of the talk.
- Treat governance and cost as ongoing work. WSO2 describes partial gateway adoption and ongoing impact measurement, not a finished operating model.
Related vault notes: Managing the Agent Lifecycle at Scale develops the lifecycle and governance side, while Building Governed Integrations and AI Agents for the Enterprise covers governed integration patterns.