🔥 Top Story
AI didn’t replace our security team — it multiplied it.
Andy Gombar, a Staff Detection and Response Engineer at Webflow, describes how the company runs security detection and response without a traditional SOC or shift-based analyst team. Instead, a small group of security engineers owns the full lifecycle — writing detections, responding to incidents, and improving the system over time. When detections jumped 200% in a single quarter, the team replaced its zero-context triage process with an AI layer that pre-assembles ownership data, correlated historical signals, and a preliminary severity score before a human ever opens the ticket, and auto-closes alerts that meet a high-confidence false-positive bar. That shift alone saved the team 504 hours in one quarter. For ambiguous investigations, engineers use LLMs to summarize raw logs, surface similar past incidents and playbooks, and draft initial timelines, while every consequential decision still requires human sign-off. After an incident closes, AI also synthesizes incident-channel messages and meeting notes into a reconstructed timeline, a post-incident analysis draft, actionable follow-up items, and updated playbooks. Gombar stresses that none of this works without clean, normalized data pipelines, an accurate asset inventory, and well-tuned detection logic already in place, and that Webflow deliberately stays model-agnostic rather than betting on a single AI vendor.
💡 Why it matters: Shows a concrete, non-vendor path to scaling incident response headcount with AI — but the payoff only materializes after data pipelines, asset inventories, and detection logic are already clean, which is the harder prerequisite most teams skip.
🔗 Read more · The New Stack
DevOps & Infrastructure
The bottleneck for AI agents isn’t the model anymore. It’s the context layer.
The New Stack
This is a contributed column by Asaf Wiener, CEO and co-founder of security startup Mate Security, arguing that AI agent reliability problems stem not from the underlying model but from a missing infrastructure layer: reliable context. Wiener writes that he's watched the same pattern repeat for two years — a team builds an agent, hits reliability problems, and reflexively upgrades to a bigger model instead of fixing the actual cause. The core claim is that dependable agents require investment in the infrastructure that feeds them trustworthy context, not a better model. Wiener previously led CSPM and vulnerability-management product areas at Wiz before founding Mate Security, so the piece is aimed at security and enterprise teams trying to run agents reliably in production. It runs in The New Stack's AI Agents / AI Infrastructure section as sponsored contributor content. Because the full article body wasn't accessible for this digest, this summary is limited to the confirmed title, description, and excerpt rather than specific techniques or figures from the piece.
💡 A useful prompt to re-examine agent reliability issues as a context/infrastructure gap rather than a model-selection problem — though as contributed content from a vendor CEO, treat specific claims as a perspective to weigh, not a vendor-neutral benchmark.
Platform engineering’s new job: serving environments at agent speed
The New Stack
This The New Stack piece, 'Platform engineering's new job: serving environments at agent speed,' argues that coding agents generate a fundamentally different kind of demand for dev environments than humans do. Human requests are diurnal and tolerant of delay, while agents ask for environments with lifetimes measured in minutes and response expectations measured in seconds, retrying and fanning out in tight loops — traffic patterns platform teams are already describing using that exact word, 'traffic.' The article notes that a 100-developer organization with agents in the loop can generate hundreds of environment requests before lunch. Its proposed fix is to stop treating environment provisioning as a one-off task and instead run it as a serving system, budgeting environment capacity the way teams already budget compute or CI runners. That reframing turns agent adoption from a surprise infrastructure bill into a plannable demand curve. Per the piece's own figures, 90% of organizations have already adopted at least one internal platform with golden paths, and the author's conclusion is that platform engineering's next job is extending a decade of self-service golden paths for developers to cover agents as well.
💡 Signals that platform teams need to budget and capacity-plan environment provisioning like a serving system — not a manual process — before agent-driven request volume turns into an unplanned infrastructure bill.
This digest was collected from RSS feeds and summarized by AI (Claude). See the original links for full details.