In April 2025, we launched Harbor Pilot with a clear mission: to empower Identity Security teams to operate as if they had a workforce ten times their actual size. Our aim was and is to transform teams of 5 into teams of 50.
The response from our customers has been overwhelmingly positive. Since the launch, we’ve achieved a 60% customer adoption rate, with retention rates nearing 80%. However, we didn’t just focus on the metrics; over the past six months, we have been actively listening to your feedback.
Your input has been invaluable. You’ve made it clear that while the concept is promising, the solution must scale to meet the complexities of enterprise needs. You requested more agents, broader domain coverage, and the capability to address intricate, multi-turn governance challenges.
To meet these demands, we are excited to introduce the next iteration of Harbor Pilot.
What’s New
We have modernized the interface to align with the best-in-class operational tools you use today.
- Real-Time “Thinking” & Streaming: We have moved away from the static “waiting” state. You will now see responses stream in real time, significantly reducing time-to-first-response.
- Visible Execution Plans: Trust is paramount in security. The interface now exposes the agent’s “thought process,” allowing you to see the logic and execution plan the AI is formulating before it acts.
- Self-Correcting: Previously, complex requests—such as intricate Admin Search queries or multi-step Workflow generation—might fail or stall. The new agents now utilize iterative reasoning to help generate better results for our users. Now we acknowledge that this won’t always work, but it represents an improvement over what exists today.
Evolving the Infrastructure
This update to Harbor Pilot is an evolution of our underlying AI engine. We have transitioned from a single-stream processing model to a robust Multi-Agent Infrastructure.
This overhaul allows us to break free from the constraints of a monolithic system, moving toward a fleet of specialized agents that can reason, plan, and execute independently while remaining tightly coordinated.
Under the Hood: The Technical Stack
For the Identity engineers and architects in our community, this release represents a shift toward vendor-agnostic, industry-standard protocols designed for extensibility:
- Orchestration via LangGraph: We have rebuilt our agents using Python and the LangGraph framework. This enables sophisticated state management, allowing agents to maintain context over long, complex troubleshooting sessions or certification campaigns.
- Agent-to-Agent (A2A) Protocol: We are adopting the A2A protocol to standardize communication between our agents. This ensures that as we add new specialized agents to the fleet, they can immediately “talk” to one another and share context without custom integration overhead.
- Model Context Protocol (MCP): To connect our agents to the real world, we are utilizing Anthropic’s MCP. Our agents now connect to a SailPoint MCP Server where our stable ISC APIs are wrapped as deterministic tools.
Why It Matters: Deterministic Outcomes
In Identity Security, “hallucination” is not an option. The primary value of this evolution is the strict separation of reasoning from execution.
By pairing non-deterministic agents (the reasoning engine) with deterministic MCP tools (the execution layer), we ensure that while the AI is flexible in understanding natural language, its actions against your environment remain strictly governed by defined, verifiable API contracts.
Future Outlook
Users will notice immediate improvements in responsiveness, including token streaming, which provides real-time feedback as the agents “think” and plan.
More importantly, this architecture lays the foundation for Canvas Mode. By standardizing on A2A and MCP, we are building a future where SailPoint agents can operate seamlessly outside the chat interface, integrating directly into your broader operational workflows.
Rollout & Availability
This infrastructure update is rolling out automatically to all ISC Suites customers through the end of calendar year 2025.
- Automatic Upgrade: The changes will automatically be turned on if you have Harbor Pilot enabled.
- How to Enable: If you are new to Harbor Pilot, Admins can enable the feature via the Global System Settings page under the GenAI tab.
Request for Feedback
Developing agentic systems for complex enterprises requires substantial effort and feedback. If you’re interested in Harbor Pilot and want to help shape SailPoint’s fleet of AI Agents, please contact me or schedule a meeting via my Calendly link.
We built Harbor Pilot to ensure our AI solutions are rooted in customer problems, not market hype. We look forward to seeing what your teams achieve with this new foundation.