Field notes on the consequences of AI: for the network, the security team, the data center and the people who answer for it. The questions worth asking before they’re obvious, with the evidence and our view.
ArticleAI assurance8 min read
Boards say they can’t see whether AI is paying off. The fix isn’t a better dashboard. It’s a different person holding the pen, and ten questions they should ask before the next renewal.
Score any AI tool, pilot or agent in 45 minutes, before you renew it, scale it or report it to the board. 12 pages.
Every piece Glappy publishes lands here, filed by month and year. Articles and field guides open on glappy.com; LinkedIn notes open on LinkedIn.
Newest first. The current month is open; older months and years fold away like a shelf.
ArticleAIInfrastructureGovernment
Rogue test agents pulled in regulators on three continents this week. What stayed with me was a money gap, a frontier model that runs on a desk, and eight states suing over kids.
By Dr. Russell Avre
Evidence: Gartner (Oct 14, 2026, forecast) · CNBC · Meta · NBC News · CERT-In
ArticleInfrastructureNetwork
It’s the cart that just learned to choose its own route. Our line: if nobody on the floor can explain why an AI-guided machine did something, it doesn’t run again until someone can.
Evidence: TCS Physical AI Readiness Report 2026, 300 manufacturers (44% unclear or no accountability for failures) · diginomica
ArticleSecurityAI
Picture your best analyst at 9:40 p.m., finishing tomorrow’s report on an AI tool she pays for herself. Banning it doesn’t stop it. It just hides it.
Evidence: Deloitte UK GenAI Workforce Survey 2026, n=25,000 (31% of GenAI users use it without their employer’s knowledge)
ArticleGovernmentAI
Public-sector confidence in AI governance is far ahead of enforcement. An unenforced policy gives leaders false cover, which may be worse than having none.
Evidence: SolarWinds public-sector survey, 810 respondents (19% of US state & local report an AI adoption framework) · EY
ArticleAISecurity
Unread AI code is fine for a button color. It is never fine for money, customer data or access. Where we’d draw the line, and why.
Evidence: BairesDev Dev Barometer Q3 2026, 705 developers (41% say virtually no AI code ships unreviewed)
ArticleAISecurity
Boards say they can’t see whether AI is paying off. The fix isn’t a better dashboard. It’s a different person holding the pen, and ten questions they should ask before the next renewal.
Evidence: PwC Directors Survey 2026 · EY AI Risk & Governance 2026 · Atlassian
Field guideAISecurity
Glappy Field Guide #1. Score any AI tool, pilot or agent in 45 minutes: the 10-line scorecard, what counts as evidence, a 30-day proof plan, a renewal gate checklist and a one-page board template.
Evidence: 12 pages · PDF · free with work email
LinkedIn noteAI
Procurement and pilots create the feeling of progress. Assurance creates the right to scale. The first ‘Bought vs Proven’ dispatch.
Evidence: Atlassian (94% use AI / 6% can scale it) · OECD Digital Government Outlook 2026
LinkedIn noteInfrastructureNetwork
South Korea is putting roughly $1B into physical AI for manufacturing. When AI leaves the laptop, the failure modes are network, safety and who is allowed to stop the line.
Evidence: Seoul Economic Daily, Korea MSIT physical-AI program (1.4131T won, 2026–2030)
LinkedIn noteAI
Individual contributors get speed. Managers inherit outcomes they may not see. Without a governed agent surface, middle management becomes the blame sink.
Evidence: Cisco MyAgent rollout (~90,000 employees) · Atlassian Agentic Pivot
LinkedIn noteGovernmentAI
AI is used in at least one area of government in 97% of OECD countries. Only 39% require a risk assessment before deployment, and 31% audit afterwards.
Evidence: OECD Digital Government Outlook 2026
LinkedIn noteAI
Code volume is rising. Review, context and governance are not. The question has moved from ‘can AI write code?’ to ‘who owns quality when it writes half of it?’
Evidence: Atlassian 2026 AI SDLC study · The Agentic Pivot
LinkedIn noteAISecurity
CIOs are starting to treat agentic AI as a new class of insider. Accountability is moving from model-safety essays to named owners across the business, not just IT.
Evidence: Fortune, 16 Sep 2026 (quoting PwC’s global chief AI officer)
LinkedIn noteAISecurity
91% of large-company AI leaders say they use agentic AI; 49% haven’t updated governance for it, and 26% can’t detect unauthorized agents at all.
Evidence: EY US AI Risk and Governance Survey, 15 Sep 2026 (n=202)