In partnership with
Hello from The AI Night,
Today in AI:
OpenAI Launches Form-Filling Feature in ChatGPT
xAI Launches Grok Integration for OpenCode
NVIDIA Verified Agent Skills Bring Capability Governance
Here's the deal: OpenAI showed ChatGPT completing paperwork by combining image uploads with voice mode. You upload a photo of a form, tell ChatGPT what to enter, and it returns a filled-out version. The feature also works through text alone.
The Breakdown:
Upload a form image and say what to fill in by voice, then get back a completed version.
It also works text-only: upload the form, add details, and ChatGPT completes it.
A demo filled a gym membership form by voice; a separate text example completed a job application.
Output is returned as a static image, not an editable PDF or interactive document.
Uploads must be clear and readable to avoid extraction errors.
The capability is rolling out to ChatGPT Plus users.
The bigger picture: Nobody discusses form-filling as an AI use case, which is exactly why it matters. This targets billions of people who never write a prompt longer than one sentence. Static image output is a real limitation. But OpenAI is not competing with Jotform. It is competing with the fifteen minutes someone spends retyping the same address into the same boxes every week.
Here's the deal: xAI now lets developers use Grok inside OpenCode, the open-source terminal coding agent. Subscribers connect a SuperGrok or X Premium account and code with Grok Build, the same model that powers xAI's terminal coding agent.
The Breakdown:
Access comes through an existing SuperGrok or X Premium subscription, with no separate API key required.
Setup installs OpenCode via a single curl command, then launches with the opencode command.
Inside OpenCode, the /connect command lets users pick xAI as the provider.
Two OAuth sign-in methods are offered: a browser flow for local machines and a headless flow for SSH or VPS hosts.
The headless option prints a code and URL, supporting remote and server-based development.
xAI says more open-source agents and integrations are coming soon.
The bigger picture: Third open-source agent integration in weeks, after Hermes and OpenClaw. The playbook is clear. OpenAI and Anthropic build the terminal and own the experience. xAI plugs into whatever developers already use. No API key, no switching cost. Every integration is a bet that the model matters more than the product wrapped around it. One of these strategies is wrong.
Here's the deal: NVIDIA introduced verified agent skills, adding provenance, security scanning, and cryptographic signing to portable agent instruction sets. The skills live in the public NVIDIA/skills GitHub repo and build on the agentskills.io open spec, so one SKILL.md runs across Claude Code, Codex, and Cursor.
The Breakdown:
Verified skills are cataloged, scanned, signed with a detached skill.oms.sig, and shipped with a machine-readable skill card.
SkillSpector scans for software risks like vulnerable dependencies, credential access, and exfiltration paths before publication.
It also flags agent-native risks: hidden instructions, prompt injection, trigger abuse, excessive agency, and tool poisoning.
Signing uses OpenSSF Model Signing, covering every file so developers verify a skill is unchanged after download.
Scanning coverage maps to OWASP LLM guidance and MITRE ATLAS; NVIDIA also released a card template and generator.
The bigger picture: Agent skills are the new npm packages. Developers will install thousands without reading every line. That is exactly how supply-chain attacks work. NVIDIA is building the security layer before the first major incident, not after. One signature check replaces hours of manual auditing. Whoever sets the trust standard for agent skills decides what runs inside every production agent.
2026 State of AEO Report
A year ago, most marketers weren't thinking about AI search. Now it's one of the fastest moving channels in the industry and nobody has a playbook yet.
So we built one. We surveyed hundreds of marketers to find out how they're approaching answer engine optimization, where they're investing, what's actually working, and what isn't.
The result is the 2026 State of AEO Report. Real data. Real strategies. A clear picture of where AI search is headed and how to get ahead of it.
What else you need to know:
GBrain shipped v0.40.0, adding a voice agent built on Google's Gemini Live to its OpenClaw/Hermes Agent, with large context and tool use, released open source under an MIT License.
Google is expanding SynthID verification for images, video, and audio to Search and Chrome, building on 50 million uses in Gemini, and launched an AI Content Detection API on Cloud.
NVIDIA released an open-source agent skill for its AI-Q blueprint, letting harnesses like Claude Code and Codex delegate research tasks to a local or hosted server for cited reports.
ElevenLabs reported its Voice Marketplace creators have earned over $22 million, doubling from $11 million in November, with more than 10,400 actors licensing voices for ongoing per-use payouts.
Grok added connectors for Vercel, Canva, Gamma, and S&P Global, letting users build and deploy sites, design assets, generate decks, and pull live market data without leaving the platform.
That’s it for today’s edition of The AI Night.
Our goal is to cut through the noise, surface what actually changed, and explain why it matters.
2 ways to support us:
Forward this to your AI-curious friend → https://www.theainight.com
Sponsor The AI Night and reach 500+ AI builders daily → passionfroot.me/theainight