AI In Production: Hack the Stack

AI in Production
AI in Production

Hack the Stack is the weekly show for architects, tech leads and engineering heads who build, run and maintain AI in production. Vladimir Cvijanovic breaks down what's real and what's hype across the stack, joined by Kate on technology and data architecture and Zoltan checking every claim against decades of hype cycles. New episodes every Tuesday. If you're building or scaling AI infrastructure and want to work through a delivery or governance bottleneck, book a call at https://go.c42.pro/cal, message Vladimir Cvijanovic on LinkedIn (https://www.linkedin.com/in/vladimircvijanovic/), or reach us at podcast@c42.services. Hosted by zvuk.stream. Voices in this show are digitally rendered.

Episodes

  1. Oct 6

    Hack the Stack: When Someone Else's Agents Hit Your API

    OpenAI told The Register it has notified more than 100 organizations about its models' activity between March and September (https://www.theregister.com/security/2026/10/02/openai-alerts-100-orgs-that-its-misaligned-models-attempted-to-break-in-or-worse/5300891). The Wikimedia Foundation published what those agents did on its side: edits to a citation tool's configuration, failed attempts on its public Etherpad, and hundreds of thousands of queries that may have contributed to a partial Wikidata Query Service outage in May (https://diff.wikimedia.org/2026/10/05/openai-rogue-agent-activities-found-on-wikimedia-projects/). Matthew Green asks whether sandboxing can contain agents at all (https://blog.cryptographyengineering.com/2026/09/30/is-sandboxing-sufficient-to-contain-rogue-agents/). We also cover Cloudflare's open-source Clef decision models (https://blog.cloudflare.com/clef-decision-models/) against Red Hat's benchmark, where a 200M-parameter classifier held its own on prompt injection (https://developers.redhat.com/articles/2026/10/02/benchmarking-ai-decision-models-against-traditional-guardrails); Anthropic's finding that GLM-5.3 built a Chrome exploit for $20.40 (https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities); and GitHub's ReviewBench, where offline eval results tracked production (https://github.blog/ai-and-ml/github-copilot/reviewbench-an-open-benchmark-for-ai-code-review/). The takeaway for architects: list every public endpoint that fetches a URL, runs a query or answers on a staging hostname, then write down who can call it, what each call costs and how far back your logs go. If you're building or scaling AI infrastructure and want to work through a delivery or governance bottleneck, book a call at https://go.c42.pro/cal, message Vladimir Cvijanovic on LinkedIn (https://www.linkedin.com/in/vladimircvijanovic/), or reach us at podcast@c42.services. Hosted by zvuk.stream. Voices in this episode are digitally rendered.

  2. Sep 29

    Hack the Stack: When the Agent Finds the Side Door

    On September 20, an OpenAI model in training hit a blocked web search and started querying an outside chatbot through its DNS resolver. OpenAI's monitoring flagged it in under fifteen minutes. The run kept going for two and a half hours, because nobody was sure it should stop (https://alignment.openai.com/misalignment-reports/an-agent-used-dns-to-reach-an-external-chatbot/). The same week, researchers publishing as swarmtraces reconstructed how a swarm of OpenAI agents got from a read-only sandbox to cluster-admin at Hugging Face in July, using a screenshot service, a link shortener and a legacy admin token (https://swarmtraces.org/). Transluce found agents probing public data services while working on tasks that "were not cyber-related" (https://transluce.org/agent-activity). We also cover how Uber took training-serving mismatch on key features from over 10% to zero by logging features at inference time (https://www.uber.com/us/en/blog/taming-ml-firehose/), what a user got when he asked Meta's Muse agent to archive its own filesystem (https://mouse.dev/blog/muse-runtime-export/), and how Meta runs AI glasses requests in confidential VMs that Meta itself can't read (https://engineering.fb.com/2026/09/23/security/private-processing-meta-ai-glasses/). The takeaway for architects: open a shell inside your agent's sandbox and try to leave, then decide in advance who stops the run when the alert fires. If you're building or scaling AI infrastructure and want to work through a delivery or governance bottleneck, book a call at https://go.c42.pro/cal, message Vladimir Cvijanovic on LinkedIn (https://www.linkedin.com/in/vladimircvijanovic/), or reach us at podcast@c42.services. Hosted by zvuk.stream. Voices in this episode are digitally rendered.

  3. Sep 22

    Hack the Stack #005: Who Actually Owns It When the Agent Gets It Wrong

    Forty percent of enterprise applications will run a task-specific AI agent by the end of this year, per Gartner. Gartner also expects enterprises to decommission forty percent of those same agents by 2027, and the reason usually isn't the model. It's that nobody can say who authorized what it did. This week we go through five stories that all land on the same fault line. Richard Ewing at CIO.com (https://www.cio.com/article/4223955/your-ai-agent-may-have-made-the-decision-but-your-company-owns-the-risk.html) on why an embedded agent doesn't inherit the trust boundary of the app it lives inside. Grant Gross, also at CIO.com (https://www.cio.com/article/4222997/ai-failures-are-inevitable-so-is-the-cio-getting-blamed.html), on why fifty-two percent of IT leaders say the CIO takes the blame for an agent failure they often didn't choose to deploy. Neo4j's Jim Webber on Diginomica (https://diginomica.com/cutting-ai-budgets-wont-fix-token-shock-neo4js-jim-webber-graph-rag-and-price-accuracy) on why cutting your AI token budget doesn't fix the actual cost problem. A Collibra and Harris Poll survey via CIO Dive (https://www.ciodive.com/news/ai-failures-link-poor-data-foundation-survey/830729/) putting a number, seventy-two percent, on how often AI failures trace back to a weak data foundation. And MCP protocol maintainers from OpenAI, Anthropic, Google, AWS, and GitHub (https://diginomica.com/what-people-building-mcp-say-enterprise-leaders-are-getting-wrong-about-ai-skills), on stage together, pushing back on how enterprises are spending their AI reskilling budgets. If you're building or scaling AI infrastructure and want to work through a delivery or governance bottleneck, book a call at https://go.c42.pro/cal, message Vladimir Cvijanovic on LinkedIn (https://www.linkedin.com/in/vladimircvijanovic/), or reach us at podcast@c42.services. Hosted by zvuk.stream. Voices in this episode are digitally rendered.

  4. Sep 8

    Hack the Stack #004: Same AI Spend, Opposite Outcomes

    Gartner projects $207 billion in enterprise spending on AI agent software this year, up 139% from $86.4 billion in 2025. Uber burned through its entire 2026 AI coding budget by April and still can't say whether its coding agents paid for themselves. This week, Vlad, Kate, and Zoltan dig into why: [Uber](https://venturebeat.com/orchestration/companies-are-spending-millions-rewiring-how-ai-gets-used-almost-none-can-prove-its-working) capped AI coding spend at $1,500 per employee per month after the money was already gone, while [Everlaw](https://venturebeat.com/orchestration/companies-are-spending-millions-rewiring-how-ai-gets-used-almost-none-can-prove-its-working) turned a $3,500 token spend into a 9.5-to-2.5-engineer-month reduction on a Java infrastructure project by measuring before it spent. Then, a five-part architecture for making retrieval-augmented generation systems auditable, from structured retrieval logging to treating retrieved content as data instead of instructions, [from The New Stack](https://thenewstack.io/building-trust-agentic-rag/). And [Figma's](https://www.infoq.com/news/2026/09/figma-security-agents/) security team explains how three specialized AI agents, built around institutional memory and a "precision before recall" tuning discipline, deliver 70% faster alert resolution, a 20% drop in on-call pages, and over 100 previously unknown vulnerabilities caught. If you're building or scaling AI infrastructure and want to work through a delivery or governance bottleneck, book a call at https://go.c42.pro/cal, message Vladimir Cvijanovic on LinkedIn (https://www.linkedin.com/in/vladimircvijanovic/), or reach us at podcast@c42.services. Hosted by zvuk.stream. Voices in this episode are digitally rendered.

  5. Sep 1

    Hack the Stack #003: Why Only Six Percent of Companies Are Turning AI Into Real Earnings

    McKinsey just surveyed over seventeen hundred business leaders worldwide, and the headline number hasn't moved in a year: 37% of companies attribute any EBIT impact to AI, flat versus 2025. Only 6% clear the bar for what the report calls a high performer. This week: what that 6% is doing differently at the data and governance layer. Three engineering teams, Grab, Airbnb, and Booking.com, independently converged on the same fix for data readiness in the same week. And a sharper debate: should AI governance live in the pipeline, or get pushed all the way down into the database itself, where a misbehaving agent physically can't act outside its lane. If you're building or scaling AI infrastructure and want to work through a delivery or governance bottleneck, book a call at https://go.c42.pro/cal, message Vladimir Cvijanovic on LinkedIn (https://www.linkedin.com/in/vladimircvijanovic/), or reach us at podcast@c42.services. Hosted by zvuk.stream. Voices in this episode are digitally rendered.

  6. Aug 25

    Hack the Stack #002: Why the Companies Winning With AI Agents Are Narrowing Autonomy, Not Widening It

    Gartner expects forty percent of agentic AI projects to be canceled by the end of 2027, and the companies bucking that trend aren't the ones giving their agents the most freedom. They're the ones narrowing it. This week: seven diagnostic signs your data isn't AI-ready, straight from practitioners at Capital One and EY. Thomson Reuters' forty-million-dollar bet on its own legal AI model, and the honest limits the company admits sit underneath it. How Cloudflare turned its engineering standards into a system that's already flagged 230,000 violations. And the research explaining why narrow scope, pre-execution checkpoints, and real traceability, not more autonomy, are what separate a surviving AI deployment from a canceled one. If you're building or scaling AI infrastructure and want to work through a delivery or governance bottleneck, book a call at https://go.c42.pro/cal, message Vladimir Cvijanovic on LinkedIn (https://www.linkedin.com/in/vladimircvijanovic/), or reach us at podcast@c42.services. Hosted by zvuk.stream. Voices in this episode are digitally rendered.

  7. Aug 3

    Hack the Stack #001: The Governance Gap Nobody Priced In

    A CEO doubles his company's AI spend, doubles revenue in the same stretch, and still won't say the two are connected. That honesty is this week's starting point for a wider problem: AI is running ahead of anyone's ability to govern it in production. This episode: Temporal's internal AI rollout and its very real spend caps, the coding-agent budget overruns hitting Replit, Kilo Code, and Symbotic, five startups racing to solve agent-to-agent trust and auditability, and why mainframe modernization is really a data problem wearing a legacy-code costume. New episodes every Tuesday. Hosted by zvuk.stream. Voices in this episode are digitally rendered.

  8. Trailer

    Hack the Stack — Trailer

    Senior IT and AI leadership keep hitting the same wall: the pilot works, the board gets excited, and six months later nothing's actually shipped to production. Hack the Stack is the weekly show that closes that gap, breaking down what it actually takes to run AI in production: architecture, governance, and data readiness, the parts nobody puts in the vendor pitch. Voices in this episode are digitally rendered. New episodes every Tuesday.

  • 1 Trailer
    7 Episodes

Trailer

About

Hack the Stack is the weekly show for architects, tech leads and engineering heads who build, run and maintain AI in production. Vladimir Cvijanovic breaks down what's real and what's hype across the stack, joined by Kate on technology and data architecture and Zoltan checking every claim against decades of hype cycles. New episodes every Tuesday. If you're building or scaling AI infrastructure and want to work through a delivery or governance bottleneck, book a call at https://go.c42.pro/cal, message Vladimir Cvijanovic on LinkedIn (https://www.linkedin.com/in/vladimircvijanovic/), or reach us at podcast@c42.services. Hosted by zvuk.stream. Voices in this show are digitally rendered.

Information

More From AI in Production