About This Episode
In this episode of Stratola Spectrum, Dinesh Chandrasekhar, Chief Analyst, Stratola, explores a problem that refuses to go away: integration.
Data integration, application integration, API integration, workflow integration. Enterprises have been fighting this battle for decades using middleware, ESBs, event brokers, connectors, and APIs. Yet two decades later, the core reality is still the same: systems remain fragmented, and getting the right data to the right place at the right time is still hard.
What has changed is the context.
In the age of AI, enterprises are not only trying to integrate systems. They are trying to integrate agents. Every major platform is building its own “agent ecosystem” inside its own walls, creating what Dinesh calls agentic bubbles. The moment you try to make these bubbles work together across systems, the old integration problem returns, just wearing a new outfit.
To unpack what this means and what a real solution could look like, Dinesh speaks with Ed Funnekotter, Chief AI Officer at Solace. Ed’s career at Solace spans 21 years, from building FPGA-based high-speed event-broker hardware to leading AI product development today. That combination gives him a rare vantage point: he understands both the deep plumbing of event-driven architecture and the messy realities of enterprise adoption.
Ed introduces Solace’s concept of an “agent mesh”: an always-on orchestration layer where multiple specialized agents can communicate, coordinate tasks, access tools (APIs, MCP servers, databases), and produce outputs with traceability. The conversation digs into what agent orchestration really requires, why the AI part is often the smaller slice of the work, and why trust, data lineage, and PII handling determine whether any of this can move beyond pilots.
They also get into the hardest parts: cascading errors in multi-agent workflows, hallucinations, evaluation strategies (including LLM-as-judge and parallel reasoning), and why enterprise-grade agent systems will likely evolve first in internal-facing workflows like support triage, reporting, and knowledge retrieval before they fully autonomize customer-facing decisions.
If you are building agentic systems inside an enterprise, this episode is a practical reminder that AI is not the main blocker. Integration, governance, and trust are.
Key Takeaways
Integration is still the real bottleneck, not the LLM. AI is often only 10 to 20% of the work. The hard part is connecting data sources, enforcing security, and making workflows reliable in production.
Agentic bubbles are real and they do not naturally interoperate. SAP, Salesforce, GitHub, Jira, and others can build great agents inside their own ecosystems. Cross-system enterprise workflows still need a strong orchestration and integration layer.
Solace’s “agent mesh” is positioned as that orchestration layer. It uses an orchestrator agent plus specialist agents to pull from multiple tools and systems. The goal is to coordinate retrieval, reasoning, and actions across systems using event driven patterns.
Trust is the blocker for production autonomy, so lineage matters. If you cannot trace why the agent gave an answer, you cannot safely deploy it. Their approach emphasizes artifacts, metadata, and source tracking so humans can validate outputs quickly.
Privacy and governance need to be left-shifted into the pipeline. PII should be cleaned before it enters long-term knowledge stores like vector databases. Agents can help summarize and retain what matters while stripping sensitive data.