1Password for Builders

What we learned: The agent's output ceiling is the quality of its input. Human ticket qualification can't be delegated. Narrow skills beat broad ones. One "implement a design system ticket" skill didn't work. Eight focused skills did. Treat agent credentials like machine credentials. Short-lived, scoped access is the baseline. Full write-up: bit.ly/3S5SMy3

The unexpected outcome: designers got the same setup. Paste a Figma link, describe intent, get an interactive prototype built from real components with a shareable deploy link for stakeholder review. The pipeline we built for engineers became a prototyping tool for the whole product team.

We pointed a coding agent at our design system backlog. The first attempt was a mess. The agent filled knowledge gaps with confident-sounding guesses. Components that looked right but were semantically wrong. PRs that passed tests but weren't idiomatic.

2/3 of organizations have already experienced a security incident tied to AI agent deployment. The problem isn’t the agents. It’s how agents are accessing credentials. @Forbes covered our collaboration with @OpenAI to bring agentic security to Codex. Dive in: bit.ly/4f1kOo5 #AgenticAI #IdentitySecurity #OpenAI #DevSecurity #1Password

Workload auth and human attribution. If you can't trace an agent action back to a human, governance is already gone. @ithilgore (@OpenAI), @JeffMalnick and @NancyZWang (@1Password) discuss why authorization frameworks weren’t built for agent identities. Zero-Shot Learning Ep 1: youtu.be/w2TFGN1IfIM?si…

If a credential reaches an agent’s context window, consider it compromised. Prompt injection doesn't need to defeat authentication. It just needs an agent that's already inside. @ithilgore (@OpenAI) unpacks the risks with @JeffMalnick and @NancyZWang (@1Password) on Zero-Shot Learning Ep 1 youtu.be/w2TFGN1IfIM?si…

Security teams 12 months ago: "Block ChatGPT." Security teams today: "Here's godlike admin privileges, let's run wild." Nancy Wang, CTO, @1Password | #WebSummitVancouver

OAuth was designed for: human clicks authorize, token issued, done. Agents start with one task and evolve mid-run based on what they find. The authorization you granted at step one may not still be appropriate by step seven. That's the gap @ithilgore (@OpenAI) and @JeffMalnick (@1Password) get into with @NancyZWang on Zero-Shot Learning Ep 1 youtu.be/w2TFGN1IfIM?si…

Your agent summarized the emails, then decided to post them to Slack. Nobody asked it to, but it thought it was being helpful. This is the problem that better prompts don't fix. You need authorization at the action level, not just the task level. bit.ly/4dByow5 @ithilgore, @OpenAI, on Zero-Shot Learning Ep 1 with @JeffMalnick and @NancyZWang from @1Password.

"As developers bring coding agents into real software workflows, secure access to credentials is critical. 1Password's MCP server for Codex helps teams give agents the access they need at runtime, without copying credentials into prompts, local files, or repositories. That’s the kind of security that simplifies agentic development, empowering teams to ship faster while keeping sensitive credentials protected.” - Nick Steele, Agent Security at @OpenAI. How it works in practice: bit.ly/4tOwzlv @OpenAIDevs #AISecurity #IdentitySecurity #OpenAI #Codex #1Password #AI #DevSecOps #Developers

Proud to see our work on agent security @openai highlighted in Forbes. Securing AI agents means bringing identity, credential, and access controls directly into the developer workflow, and Codex is a major step in that direction. forbes.com/sites/timkeary…

Toronto-based @1Password has expanded its work with Silicon Valley AI giant @OpenAI in an attempt to help businesses deploy autonomous AI agents to securely build products. betakit.com/openai-partner…

With @1Password Environments MCP Server for Codex, developers can: ✅ Catch secrets in plain text, move them into 1Password, and replace them with references in code ✅ Reference vaulted secrets inside Codex without the values appearing in code, terminals, or model context ✅ Replace every hardcoded credential with a vaulted reference #AISecurity #IdentitySecurity #OpenAI #Codex #1Password #AI #DevSecOps #Developers

Today, we're announcing an expanded collaboration with @OpenAI to bring trusted, just-in-time credential access to Codex. Instead of putting credentials directly into prompts or files, we provision a secure runtime environment where secrets are mounted, used, and discarded, with user authentication required at the moment of access. Learn more: bit.ly/4fv5YWW #AISecurity #IdentitySecurity #OpenAI #Codex #1Password #AI #DevSecOps #Developers

Zero-Shot Learning is here, hosted by @1Password CTO @NancyZWang and Dev Tagare, @Google’s Head of Engineering @GeminiApp. Most AI podcasts feature people talking about AI. Zero-Shot Learning features the people building it. Season 1 features guests from @OpenAI, @cursor_ai, @braintrust, @vercel, and more on the hard problems that don’t have clean answers yet. 🎧 Lock in: bit.ly/4dz8kSj #ZeroShotLearning #AI #Podcast #AIAgents #Cybersecurity #1Password

"Without attribution, we lose governance." - @ithilgore Agent Security Lead @OpenAI. When an agent acts across tools, sessions, and systems, can you prove what happened and who was responsible? This conversation on identity and access controls for agents is a must-listen. 🎙️ Zero-Shot Learning, Episode 1 drops tomorrow. #ZeroShotLearning #AIAgents #Cybersecurity #AI #Podcast

There's no shortage of AI hype. Zero-Shot Learning is the podcast for AI builders who want to go beyond it. Hosted by @1Password CTO @NancyZWang and @Google's Dev Tagare, each episode goes deep with builders from @OpenAI, @cursor_ai, @Braintrust, @Vercel, and more on the workflows, tools, and ideas shaping AI today. 🎧 Episode 1 coming soon. Lock in and stay with us for the season. #ZeroShotLearning #AI #1Password #Podcast

On the AI Summit stage @WebSummit Vancouver, @NancyZWang joined @scribeceo, @YayLT, CEO at @Webflow, and @johnkoetsier, columnist @Forbes, to address the gap between model performance and what organizations are actually getting. The bottleneck isn't the models. It's the missing context layer, the governance gaps, and identity challenges that traditional security frameworks weren't built for. "Agents aren't just machines. They can reason. They're non-deterministic. And so even if they have the right access, the question now becomes: are they actually going to do the right thing? Agents need their own identity and a way to bind intent to that identity so they can be monitored as they get access to sensitive systems." #WebSummit #1Password #AI #WebSummitVancouver