NSF Wants to Fund the Protocols That Make AI Agents Talk to Each Other

The race to build useful AI agents is accelerating across every sector. But there is a foundational problem that most of the hype ignores: these agents cannot reliably communicate with each other. They lack shared protocols, common security models, and interoperable standards — the kind of boring-but-essential infrastructure that makes complex systems actually work.

On February 20, NSF published a dear colleague letter that takes direct aim at this gap. The letter, issued under the Pathways to Enable Secure Open-Source Ecosystems (PESOSE) program, invites proposals focused specifically on the protocols enabling AI agent ecosystems. It is one of the first federal funding signals dedicated not to building AI agents themselves, but to building the connective tissue between them.

Why This Matters More Than Another AI Grant

Most AI funding targets capability — making models smarter, faster, or more specialized. This funding targets infrastructure. NSF is asking: what happens when thousands of AI agents built by different organizations need to coordinate on a shared task? How do you authenticate them? How do you ensure one agent's actions do not compromise another's data? How do you build this without locking everyone into a single vendor's ecosystem?

These are not hypothetical questions. Healthcare systems are deploying AI agents that need to query each other's knowledge bases while respecting patient privacy regulations. Supply chain platforms are running autonomous agents that negotiate across organizational boundaries. Scientific workflows increasingly involve AI systems from multiple labs processing shared datasets. In every case, the lack of open, secure protocols is the bottleneck.

NSF's letter makes clear that the agency sees this as a national competitiveness issue, not just a technical one. The PESOSE program sits under multiple NSF directorates — CISE, ENG, BIO, EDU, GEO, MPS, and SBE — signaling that the agency views AI agent infrastructure as relevant across all of science and engineering.

What NSF Is Looking For

The dear colleague letter lays out six priority areas for proposals. Each one maps to a specific gap in the current AI agent landscape.

Interoperability is the headline. NSF wants proposals that develop open standards enabling agents to work across platforms and organizations without custom connectors. If you have been building protocol-level infrastructure for multi-agent systems — message formats, discovery mechanisms, capability negotiation — this is your moment.

Scalability comes next. The agency wants architectures that support large networks of agents with straightforward integration paths. Proposals should address what happens when agent ecosystems grow from dozens to thousands of participants.

Security gets the most detailed treatment. NSF specifically calls out cross-domain data classification, identity verification, role-based access control, secure communication channels, authentication frameworks, zero-knowledge proofs, auditability, and formally specified message formats. This is an unusually specific wish list for a dear colleague letter, and it tells you exactly what reviewers will prioritize.

Open science applications must demonstrate how these security features enable rather than restrict open research. NSF does not want protocols that lock down data — it wants protocols that make sharing safe enough to be the default.

Partnerships across academic, nonprofit, industry, and government sectors are explicitly encouraged. Solo-PI proposals will likely be at a disadvantage.

Education and training must be woven in. NSF wants proposals that expand training for students and postdoctoral researchers in AI agent ecosystem development, with emphasis on security and reliability skills.

How to Apply

Proposals should follow the full PESOSE solicitation (NSF 26-506) and use the title format "PESOSE / AI: Track #: (title)." The letter notes that all research must be unclassified and publicly releasable. Questions can be directed to PESOSE@nsf.gov.

The PESOSE program itself is not new — it has funded open-source ecosystem development for several years. But this dear colleague letter represents a targeted pivot toward AI agent infrastructure, and it comes at a moment when the commercial sector is pouring billions into proprietary agent frameworks. NSF is betting that open-source alternatives will matter for science, security, and long-term competition.

Who Should Apply

The sweet spot for this funding is researchers working at the intersection of distributed systems, security, and AI — people building protocols rather than models. Specific communities that should pay attention include those working on multi-agent system architectures, open-source AI toolchains, federated learning infrastructure, secure computation protocols, and formal verification of AI system behaviors.

Computer science departments with strengths in networking, security, or distributed systems should look at this closely. So should researchers in any scientific domain where multi-agent AI workflows are becoming standard — genomics, climate modeling, materials science, and astronomy all qualify.

Small businesses and startups building open-source AI infrastructure should explore the PESOSE program's industry partnership mechanisms. NSF is explicitly looking for cross-sector collaboration, and a university-industry partnership that addresses both the research and adoption sides of the problem will be compelling.

The Strategic Context

This dear colleague letter did not emerge in a vacuum. The Model Context Protocol, Anthropic's open standard for connecting AI systems to tools and data sources, launched in late 2024 and has gained rapid adoption. Google's Agent-to-Agent (A2A) protocol followed. OpenAI has its own agent infrastructure play. The commercial ecosystem is fragmenting into competing standards.

NSF's move signals that the federal government sees interoperability as a public good — something that cannot be left entirely to market dynamics. For researchers positioned at this intersection, the timing is excellent. The need is obvious, the commercial activity creates urgency, and the funding signal is clear.

If your work touches AI agent communication, security, or open-source infrastructure, start drafting now — Granted can help you identify the right PESOSE track and build a proposal that matches what NSF reviewers will be looking for.