For the past six months, I have observed how various organizations in Australia are engaging with AI.
I’ve seen:
- What’s working.
- What’s failing.
- What nobody is discussing.
In response, we have launched a podcast.
Introducing Agents After Dark by Prefactor. This show is tailored for engineering, product, security, and AI leaders who are building real AI systems.
Expect:
- No hype.
- No speeches about how "AI will change everything."
- No recycled LinkedIn commentary.
Instead, we offer honest conversations with those deploying AI within large enterprises, technology companies, and startups.
Our initial guests include leaders from organizations managing billions of dollars, creating AI products utilized by thousands of customers, and addressing the challenges of transitioning AI from experimentation to production.
Topics we cover include:
- AI agents in production
- Quality and reliability
- Risk and trust
- Lessons from real deployments
- What successful teams are doing differently
If you are involved in building, deploying, or managing AI systems, you may find our content valuable.
The first two episodes are now live on Spotify, YouTube, and Apple Podcasts. Links in the comments.
@prefactordev@MrTeale@joshgillies
Healthcare governance optimizes for legal defense, not patient outcomes. Teams avoid deploying beneficial AI to minimize exposure instead of deploying it safely. That misalignment gets patients hurt worse than the AI ever could.
Your monitoring stack was built for humans watching services. Agent crashes happen silently at machine speed. OS-level failures, context overflow, checkpoint bloat—none hit your alerts until users complain things are broken.
Teams deploying agents to production with zero spend controls. $47K in 11 days from ping-pong loops. $400 weekend spikes from retries. Building internal proxies just to track what's happening. This is what happens when you ship capability without governance.
Defensive agents reasoning faster than humans just moved from aspirational to mandatory. SANS leadership: only viable architecture going forward. The automation gap is real and widening. Governance infrastructure can't lag behind defensive velocity. @onersac#AI#Security#Ent...
AI weaponization across attack vectors isn't theoretical anymore. Ed Skoudis just confirmed every threat surface is compromised. Defense infrastructure lags capability. Governance has to move faster. @onersac@HCAHealthcare@CVSHealth#AI#Security#RSAC
Vendors shipping agentic threat detection at RSAC but nobody's asking: who governs the AI making kill decisions on your network? Autonomous remediation without auditability is just faster breaches. @onersac#AI#Cybersecurity#Governance
Users hate black boxes. Developers love them. Enterprise adoption stalls when you can't explain why the agent did what it did. Transparency infrastructure isn't optional—it's the only bridge between capability and trust. @onersac#AI#Governance#Enterprise
Two hours. That's how long before one agent compromised another's full infrastructure. Agent-on-agent attacks are the new supply chain risk. When AI moves faster than incident response, you need autonomous defense or you're already breached. @onersac#AI#Security#Governance
Agents inherit human trust without human judgment. GitHub issue prompt injection poisoned 4,000 developer machines because the agent looked legitimate. System privileges + zero intent validation = catastrophe. Runtime governance isn't optional. @onersac#RSAC#AgentSecurity
Agent identity is the security primitive enterprises missed. Hardware-backed runtime governance means auditable agent behavior from first execution. Policy theater dies when regulators can actually verify what agents do. @onersac week confirms it: governance infrastructure win...
73% of enterprises deployed agents. 7% have real-time governance. That 66-point gap isn't technical debt—it's organizational liability. Agents rewriting their own policies demand operational visibility before velocity. @onersac gets it. @HCAHealthcare@CVSHealth@Cigna need it...
Developers ship surprise-me agents. Users want buttons that do exactly what they clicked. This gap kills adoption faster than any technical limit. Governance infrastructure that makes agent behavior predictable and explainable wins the market. @onersac#AI#Governance#Enterprise
Pilot-to-production gap isn't a tool problem. Enterprises built data, APIs, edge compute for human workflows. Agents demand different infrastructure entirely. @onersac speakers nailing it: readiness beats tool selection. #AgenticAI#Enterprise
Intent monitoring without competence verification is security theater. Three platforms at @onersac launching agent guardrails but none can actually evaluate what agents know how to do. You're governing blind. #RSAC#AgenticAI#Enterprise
Two hours. That's how long it took an AI agent to own McKinsey's chatbot with full read-write access. This isn't a security failure—it's the predictable outcome of deploying agents without governance infrastructure. Agent-versus-agent warfare is happening in production right n...
Enterprises deployed non-human identities that outnumber humans but can't audit a single action. That's not innovation—that's regulatory exposure dressed as efficiency. RSAC consensus: governance infrastructure moves now or auditors move in. @onersac#AgenticAI#Enterprise