NSF Targets AI Agent Ecosystems With New Open-Source Funding Push

The National Science Foundation fired a starting gun on February 20 for researchers building the plumbing beneath the AI agent boom. A new Dear Colleague Letter channels the agency's Pathways to Enable Secure Open-Source Ecosystems (PESOSE) program toward proposals that develop the open protocols AI agents need to communicate, collaborate, and scale.

Open Standards for a Fragmented Landscape

AI agents — autonomous software that can browse the web, write code, and orchestrate multi-step workflows — are multiplying fast. But most run on proprietary connectors. NSF's bet is that the next leap requires shared infrastructure, much like the open protocols that made the early internet possible.

PESOSE will prioritize proposals addressing interoperability (open standards so agents built by different organizations work together without custom wiring), scalability (architectures supporting large networks of agents, tools, and services), security (hardening agent-to-agent communication against adversarial manipulation), and education (training the workforce to build and maintain these ecosystems).

The solicitation (NSF 26-506) supports Phase I planning awards up to $300,000 over two years, and Phase II execution awards up to $750,000 over three years.

Why This Matters for Grant Seekers

PESOSE sits inside NSF's Technology, Innovation, and Partnerships (TIP) directorate — the fastest-growing corner of the agency's budget and one of the few areas the current administration has actively championed. Proposals that link AI agent protocols to national security, supply-chain resilience, or workforce development are likely to score well under the current political climate.

Researchers in computer science, systems engineering, and cybersecurity should pay particular attention. The DCL explicitly welcomes cross-sector partnerships with industry and nonprofits, making this a strong fit for university-industry consortia.

How to Position a Proposal

NSF encourages applicants to contact PESOSE program officers before submitting. Given the breadth of the AI agent space, a focused proposal — say, open standards for agent authentication or a reference architecture for multi-agent scientific workflows — will likely outperform sweeping manifestos. Tools like Granted can help teams identify complementary federal funding and sharpen their narrative before the submission window closes.