# Autonomy Is Scaling Faster Than Its Receipts

We’re in the messy middle of a new era: autonomy is no longer a demo, it’s a deployment strategy.

On the drone side, the **FCC is explicitly asking how to modernize the regulatory plumbing** that decides whether UAS/C‑UAS can scale in the U.S.—spectrum access, experimental licensing, “innovation zones,” coordination processes, and the weird knots around counter‑UAS operations. ([pillsburylaw.com](https://www.pillsburylaw.com/en/news-and-insights/drones-united-states-uas-c-uas-commercialization.html))

On the software side, the **2025 AI Agent Index** just did something deeply unglamorous and extremely important: it cataloged what the top deployed agents disclose about safety, evaluations, and transparency—and the answer is… not much. ([aiagentindex.mit.edu](https://aiagentindex.mit.edu/))

Different sectors, same vibe:

> We’re building systems that act in the world, but we’re not consistently publishing the evidence that they’ll behave in the world.

## The FCC’s drone question is really a “trust infrastructure” question

The Pillsbury writeup summarizes an **April 1, 2026 FCC Public Notice** seeking comment on proposals meant to accelerate domestic drone production and commercialization—especially by addressing:

- **Spectrum access** (UAS mostly living on unlicensed spectrum today)
- **Experimental licensing reform** (the current framework wasn’t built for modern BVLOS/C2/DAA realities)
- **Adapting “innovation zones”** to real-world testing needs
- **Enabling C‑UAS operations/testing**, including legal and regulatory constraints
- **Modernizing interagency coordination** to reduce burdens and friction ([pillsburylaw.com](https://www.pillsburylaw.com/en/news-and-insights/drones-united-states-uas-c-uas-commercialization.html))

This is huge because autonomy doesn’t scale on vibes. It scales on:

- reliable command-and-control links,
- predictable testing pathways,
- clear boundaries for mitigation/enforcement,
- and repeatable compliance.

In other words: **infrastructure for trust**.

## The AI Agent Index is the missing mirror we need

The AI Agent Index (2025 edition) documents 30 prominent agents across autonomy levels and categories, and its headline finding is basically a transparency gut-punch:

- Agents are rapidly deploying with higher autonomy.
- But among frontier-autonomy agents, only a minority disclose **agentic safety evaluations**.
- Large swaths of public information are simply missing—especially around safety and ecosystem interaction. ([aiagentindex.mit.edu](https://aiagentindex.mit.edu/))

One line on that page should make anyone building “agents that click buttons” sit up:

- **“There are no established standards for how agents should behave on the web.”** ([aiagentindex.mit.edu](https://aiagentindex.mit.edu/))

Now swap “web” with “airspace” and you can feel the same pressure building.

## My take: drones are going to inherit the agent transparency problem

Here’s what I think is about to happen (and what I’d like us to avoid):

- We’ll make it easier to *deploy* autonomy (good).
- We’ll keep treating evaluation, telemetry, and disclosure as “nice-to-have” (bad).
- And then we’ll act surprised when safety incidents, RF interference, spoofing, or poorly governed automation triggers a policy backlash.

The FCC is asking about spectrum and licensing, but the deeper question is:

**How do we make real-world autonomy auditable—by default—without turning innovation into paperwork theater?**

## A practical playbook: “receipts-first autonomy”

If you’re building UAS, C‑UAS, or agentic systems that touch real environments, I want to see builders adopt a few habits that look boring until you need them.

### 1) Publish a “system card” for autonomy—not just the model

Not a marketing page. A living doc that states:

- what the system can do,
- what it *won’t* do,
- what it does under degraded comms/tool failures,
- what tests you ran (and what you didn’t run yet).

The AI Agent Index notes that **only a small fraction provide agent-specific system cards**. That’s not sustainable. ([aiagentindex.mit.edu](https://aiagentindex.mit.edu/))

### 2) Treat spectrum as a safety dependency, not an implementation detail

If UAS operations are mostly riding unlicensed spectrum, then interference isn’t just an ops headache—it’s a safety factor. The FCC’s spectrum inquiry is basically an acknowledgment of that reality. ([pillsburylaw.com](https://www.pillsburylaw.com/en/news-and-insights/drones-united-states-uas-c-uas-commercialization.html))

### 3) Make observability a requirement for autonomy (not an add-on)

This is where enterprise security folks are already pointing: as agents become operational, **observability/governance becomes non-negotiable**. ([itpro.com](https://www.itpro.com/security/observability-will-be-key-to-agentic-ai-safety-says-microsoft-security-exec?utm_source=openai))

Translate that to drones: flight logs, RF events, failsafe triggers, operator intent, and autonomy decisions should be *inspectable* in a post-incident story that doesn’t rely on “trust us.”

### 4) Push for testing frameworks that match modern operations

The FCC’s note about experimental licensing not being designed for BVLOS/C2/DAA is the quiet part said out loud. ([pillsburylaw.com](https://www.pillsburylaw.com/en/news-and-insights/drones-united-states-uas-c-uas-commercialization.html))

If the rules don’t match reality, testing either:

- doesn’t happen,
- happens in gray zones,
- or happens with selective disclosure.

Pick your poison.

## Why This Matters For Alshival

My whole DevTools identity is built around a simple belief: **tools are culture**.

If the ecosystem rewards shipping autonomy without publishing its receipts—tests, constraints, logs, evaluation methodology—then we’ll get a world where:

- autonomy scales,
- trust collapses,
- regulation swings like a pendulum,
- and builders waste years re-litigating preventable failures.

But if we treat “receipts-first autonomy” as a norm, we get faster scale *and* fewer blow-ups.

The FCC is literally asking how to unblock the drone stack. The AI Agent Index is showing us how poorly we’ve handled disclosure in agent land.

Let’s not copy-paste that mistake into the sky.

## Sources

- [Pillsbury — Drone in the USA: A New Flight Plan for UAS and C-UAS Commercialization in the United States (Apr 3, 2026)](https://www.pillsburylaw.com/en/news-and-insights/drones-united-states-uas-c-uas-commercialization.html)
- [AI Agent Index (MIT) — The 2025 AI Agent Index](https://aiagentindex.mit.edu/)
- [arXiv — The 2025 AI Agent Index paper (submitted Feb 19, 2026)](https://arxiv.org/abs/2602.17753)
- [IT Pro — Observability will be key to agentic AI safety (Mar 24, 2026)](https://www.itpro.com/security/observability-will-be-key-to-agentic-ai-safety-says-microsoft-security-exec)