Developers
2-3 hrs
Per PR review + build investigation
Support
Hours
Digging through telemetry per incident
Product Owners
Half-day
Writing detailed PBI specs
FinOps
Weekly
Manual cost report generation
What if AI could do this work for them?
AI Assistant for Every Persona in Your Team
Developers
Support
Product Owners
Architects
FinOps
Delivery Leads
Repetitive Work Drains Every Role
2-3 hrs
Per PR review + build investigation
Hours
Digging through telemetry per incident
Half-day
Writing detailed PBI specs
Weekly
Manual cost report generation
What if AI could do this work for them?
AI That Understands Your Team
One AI assistant that knows YOUR team's context, with tailored skills for every persona, that acts on your behalf — not just chat
Understands your team's processes, ADO structure, coding standards, and workflows
Connects with ADO, Azure, Databricks, Kusto, Git — read AND write
Creates PRs, updates work items, generates reports — with your permission
Find experts, search wikis & ADO, understand team structure, contributions — guide to the right person
What Sets Triagent Apart
| Capability | Internal AI Tools (PAL, Sidekick, GPT-4o) | Triagent (Claude) |
|---|---|---|
| Can Execute Actions | ✗ Chat only — provides suggestions but user must copy/paste and run manually | ✓ Full autonomy — creates PRs, updates ADO, runs CLI commands with permission controls |
| Knows Your Team | ✗ Generic — no team-specific knowledge of your ADO, standards, or workflows | ✓ Deep integration — loads team-specific skills and context, understands repos, wiki, and processes |
| Has Compute Sandbox | ✗ No sandbox — cannot run code, execute scripts, or access file systems | ✓ Azure Container Apps — isolated container with Git, Azure CLI, Python, .NET |
| Multi-Agent Skills | ✗ Single model — one conversation, cannot delegate or coordinate subtasks | ✓ Specialized subagents — PR-reviewer, telemetry-investigator work in parallel |
| Deep Research | ✗ Limited context — cannot explore codebases, read documentation, or synthesize findings | ✓ Extended thinking — multi-step research across repos, wikis, and web with comprehensive analysis |
See Triagent in action
Presenter: Switch to Triagent Web UI
Command-line interface for technical users comfortable with terminal - runs locally with full file system access, Git integration, and direct tool execution to interact with local system files
Browser-based interface for all personas - secure sandboxed execution, team authentication, no local setup required
Azure Container Apps with instant availability
1000s of concurrent users, global deployment
Single AI Foundry API, per-team billing options
Share skills across teams and organizations
All acting on-behalf-of the user - review, investigate, fix, and create PRs autonomously
Understands the support process set by each team and follows established procedures
Automatically creates LSI tickets with proper categorization and details
Suggests fixes and creates backlog items for the development team
Complete investigation with detailed analysis
Time saved per investigation
One platform, tailored for each role
Backlog items with full test specs, unit tests, and acceptance criteria ready for execution
Feature epics auto-generated with stakeholder requirements and priority management
PR validation against design specs, comprehensive reports, and deep research
Team structure visibility, risk tracking, backlog coordination across teams
Azure cost reports on demand — daily, weekly, monthly summaries by region
Test case generation, edge case coverage, and integration test specifications
From Single Team Tool to Enterprise Platform
Triagent for Omnia Data Team
Deloitte-wide Skills & Agents
Any team can publish skills → Any team can consume
Built on Claude Agent SDK — Same foundation as Anthropic's official agent marketplace
Transforming how teams work
One Platform. Every Persona. Infinite Possibilities.
Questions?
100% Powered by Microsoft Azure
Claude Models (Opus / Sonnet)
Session Pool (Sandbox)
Chainlit Web UI