From MCP to Multi-Agent: The AI Concepts Every PS Leader Needs Right Now

Sri Ganesan's Propel 26 closing session—10 AI concepts, real examples, and one challenge: go build something this weekend.

You can't lead your team through an AI transformation you haven't personally experienced.

That was the central message from Rocketlane CEO Sri Ganesan's closing keynote at Propel 26. While many Professional Services leaders are still evaluating AI from a distance, the leaders gaining real intuition are already experimenting, building, and learning where the technology succeeds—and where it breaks.

The session wasn't a vision of some distant future. It was a practical walkthrough of ten AI concepts that PS leaders can start exploring today. 

From Model Context Protocol (MCP) and reusable AI skills to voice agents, computer use, multi-agent orchestration, and persistent memory, Sri's challenge was simple: stop watching and start building.

Read on for the key takeaways from the session.

Why PS leaders can't outsource AI experimentation

Sri opened with a confession.

He didn't build the presentation himself. He asked Claude to do it.

Not because he lacked the time, but because he wanted to force himself into the experience of building with AI rather than simply talking about it.

"I did not make this deck. I asked Claude to make it because I also need to be AI-pilled."

The lesson extends beyond slide creation.

Many leaders are asking their teams to adopt AI while remaining observers themselves. But intuition only develops through direct experience. You learn where prompting breaks down. You discover what workflows are genuinely useful. 

You start recognizing the difference between an impressive demo and a meaningful capability.

The next phase of PS leadership isn't simply understanding AI conceptually. It's understanding how work changes when AI becomes part of the execution layer.

What is MCP—and why does it matter for Professional Services?

One of the most important AI concepts discussed was MCP, or Model Context Protocol.

Sri described MCP as the connective tissue that allows AI agents to work across multiple systems. Traditional APIs require rigid integrations and tightly defined contracts. MCP introduces a more flexible model in which agents can discover and use capabilities based on context.

For Professional Services teams, the implications are significant.

An agent can pull information from your CRM, PSA, calendar, email, and project systems within a single workflow. Instead of manually moving information between tools, the agent automatically orchestrates the process.

Rocketlane has leaned heavily into this model through Nitro AI and a growing library of MCP tools built directly into the platform.

One example demonstrated during the session was an agent reading a consultant's calendar, mapping meetings to open Rocketlane tasks, and preparing timesheet entries for review. It respects company policies, validates time allocation, and waits for approval before committing updates.

The outcome isn't just convenience. It's better compliance, reduced administrative work, and more accurate delivery data.

Configuration MCP may be even more impactful.

Rocketlane's team built over 150 configuration MCP tools covering setup and administration tasks that traditionally required days or weeks of manual work. During live customer engagements, these tools produced remarkable results:

• 29 project templates configured in 2 hours instead of 2–3 weeks
• 50 template field updates completed in 5 minutes
• Seven regional holiday calendars configured in roughly 10 minutes

This is where MCP becomes transformational—not simply connecting systems but automating the operational work that slows down implementations.

Skills: turning expertise into reusable AI workflows

If MCP provides connectivity, skills provide repeatability.

Skills are essentially reusable operating procedures for AI. They capture how a task should be performed and make that knowledge available across the organization.

Think about all the repeatable work inside a PS organization:

• Kickoff preparation
• Health check reviews
• Project audits
• Scope change assessments
• Executive status reporting

Most teams rely on tribal knowledge to execute these activities consistently.

Skills change that.

Instead of relying on a single experienced consultant to remember the process, the workflow is available to everyone. Expertise scales beyond the individual.

As Sri explained, one person's best practice becomes the team's default behavior.

For growing services organizations, that's an incredibly powerful lever.

How voice agents and computer use are changing delivery workflows

Several concepts explored during the session focused on changing how work itself gets done.

Voice agents are one example.

Using tools like ElevenLabs and low-code development platforms, teams can build AI agents that conduct structured conversations with stakeholders, collect information, identify alignment gaps, and return summarized insights.

Imagine a kickoff process where key stakeholders are interviewed before the first meeting even happens. Instead of spending the kickoff gathering basic information, the implementation team begins with a synthesized context and identified areas of disagreement.

Computer-use agents solve a different problem.

Many customer environments still rely on systems with limited or no APIs. Historically, extracting information from these environments required significant manual effort.

Computer-use agents can interact directly with software interfaces the same way a human would—clicking through screens, gathering information, and performing repetitive actions.

Sri shared an example where Rocketlane used a computer-use agent to extract data from a healthcare system that lacked export functionality. Rather than manually navigating screens for hours, the agent collected and packaged the information for migration.

For implementation teams managing legacy systems, this capability can dramatically reduce migration effort and accelerate project timelines.

The future: outcome loops, multi-agent systems, and persistent memory

The final concepts Sri covered were more forward-looking but equally important.

Outcome loops allow agents to continue working until a desired objective is achieved rather than simply executing a fixed sequence of tasks.

Persistent memory enables agents to remember context across interactions and personalize future actions.

Multi-agent orchestration introduces specialized agents working together under a coordinating agent—different models handling planning, execution, research, or validation based on their strengths.

Dreaming, one of the newest concepts discussed, allows agents to automatically incorporate learnings from previous runs rather than requiring manual updates to workflows.

These aren't just theoretical concepts.

They're increasingly becoming the design patterns that AI-native platforms will adopt.

Rocketlane's Nitro AI is already evolving in this direction because it sits inside the system of record for delivery. Unlike standalone AI tools, Nitro has access to project data, resource allocations, financials, timelines, and customer context. 

That first-party visibility allows it to make more informed decisions and provide governance alongside automation.

4 key takeaways for PS leaders getting started with AI

MCP may be the highest-leverage AI infrastructure investment available today. Configuration MCP in particular can compress implementation setup work from weeks into hours.

Skills allow expertise to scale without hiring. Every repeatable process can become a reusable capability available across the organization.

Experiment with voice and computer-use agents now. The tools are accessible, inexpensive, and capable of testing real delivery workflows today.

The future of work is shifting from doing to confirming. Increasingly, AI will initiate actions while humans provide oversight, judgment, and approval.

Conclusion

Sri's challenge to the audience was refreshingly simple:

Go build something.

Not next quarter.

Not after creating an AI task force.

Not after waiting for someone else to prove the use case.

This week.

The infrastructure that felt experimental a year ago is now accessible to every Professional Services team. MCP, voice agents, computer-use systems, multi-agent orchestration, and persistent memory are no longer research projects. They're practical tools available right now.

The leaders who develop fluency with these concepts today will be far better positioned to lead the next phase of service delivery tomorrow.

The question isn't whether AI will change how Professional Services teams operate.

It will. The question is whether you'll be leading that change—or catching up to it.