Welcome to Open Channels News, your source for the latest insights in open source and digital infrastructure. In today’s episode, Robert Jacobi highlights three AI-driven stories shaping the world of technology:
Markdown for AI Efficiency Gains:
A simple formatting language is streamlining search and data inputs for AI models, saving resources and reducing costs as companies move from complex formats like JSON to well-structured Markdown 01:23. This shift can cut token usage significantly, making AI workflows faster and more affordable.
AI in Open Source Security:
AI systems are helping identify software vulnerabilities at scale, but they also risk overwhelming open source maintainers with inaccurate or duplicated reports 02:31. Effective validation and human oversight are critical to ensuring that automation saves time rather than creating more work.
The Hidden Costs of AI Agents:
Autonomous AI agents are impressive, but their operating costs can quickly outpace the value they provide 03:45. Every action, from running tools to looping tasks, adds expense. The key is measuring results against costs and ensuring that efficiency is more than just activity volume.
Our sponsors keep the lights on.
Take a moment to check out our current sponsors.
Since 2005, Automattic has built tools for the open web including WordPress.com, WooCommerce, and Jetpack that are used by millions of people to create, sell, and publish online. They believe in ownership, flexibility, and open source and we’re grateful for their support. Learn more at automattic.com.
In five days, Omnisend moves every flow, list, and template off your current platform, and you could be paying up to 35% less without lifting a finger. You just show up when it’s done. We’re glad to have them supporting the show and the community we’re building around it. Use the code OpenChannels and get 30% off your first 3 months of any paid plan.
Takeaways
- Markdown for AI Efficiency: Robert explained that using Markdown instead of JSON for feeding data to AI can significantly reduce token usage and costs, making it a more efficient format when the next consumer is an AI model needing context 01:23.
- AI in Open-Source Security: Robert highlighted that while AI makes it easier to find potential vulnerabilities in open-source projects, it also increases the risk of overwhelming maintainers with inaccurate or duplicate reports, making validation crucial 02:31.
- Validation Over Volume: Robert emphasized that automating security processes should focus on confirming real vulnerabilities and not just increasing the number of findings, thus saving valuable engineering time 03:23.
- Economic Realities of AI Agents: Robert discussed that the real value in autonomous AI agents comes only if the outcome justifies the cost, and that agents performing tasks inefficiently can quickly become economically unsustainable 03:45.
- Automation with Purpose: Robert concluded that automation in AI should always be measured by useful output, not just activity or speed, applying this principle to data formatting, cybersecurity workflows, and the economics of autonomous agents 04:47.
Mentioned Links and Resources
- The Next Web (Markdown’s Growing Role in AI Data Formatting) – The Next Web reports that Markdown is increasingly useful for passing search and other data to AI models.
🔗 https://thenextweb.com/news/markdown-default-language-search-data-ai-models - SerpApi (API Markdown Output Example) – SerpApi introduced Markdown output across more than 100 APIs, with token savings averaging around 50% and up to 90% on some endpoints.
🔗 https://serpapi.com/ - TechRadar (Open-Source Cybersecurity Vulnerabilities) – TechRadar discusses the increased vulnerability reporting in open-source cybersecurity due to AI. (Note: this is my best match among TechRadar’s pieces on the topic. It covers AI as both a problem and a solution for open source security. The original audio may help confirm the exact article.)
🔗 https://www.techradar.com/pro/security/in-open-source-cybersecurity-ai-is-kind-of-a-problem-but-it-can-also-be-a-solution - Fungus: Agentic Costs Are Unsustainable (AI Agent Economics Analysis) – Fungus explores the high and sometimes unsustainable costs associated with autonomous AI agents.
🔗 https://fungus.substack.com/p/agentic-costs-are-unsustainable
Episode Transcript
Robert Jacobi: Hello and welcome to Open Channels News. I’m your host, Robert Jacobi. Last week we were at Open Source Summit Europe, and we’ll have follow-up reports. This week, we will continue open-source discussions at CyberChess and Baltic Domain Days.
Today, we’re looking at 3 sides of the same AI story: how we feed information into these systems, what happens when we turn them loose on cybersecurity, and whether the economics of autonomous agents actually makes sense. Because faster isn’t automatically better, and cheaper isn’t always cheaper once the bill arrives.
First up, The Next Web reports that Markdown, yes, that little simple formatting language behind README files, is becoming an increasingly useful format for passing search and other data to AI models. The argument is pretty straightforward. Models are comfortable with headings, lists, and tables. If you can remove a bunch of structural overhead before passing information into a model, you may save tokens, time, and money. SerpApi, which contributed the article, says it’s introduced Markdown output across more than 100 APIs. In one company-reported example, a search response shrank from roughly 24,000 tokens in JSON to 6,400 in Markdown. That’s about a 74% reduction. Those are vendor figures, not a universal benchmark.
That’s an interesting result. JSON isn’t going away. If software needs a rigid schema and predictable fields, JSON is still great. But when the next consumer is an AI model needing context, plain, well-structured Markdown could be a much better fit. The lesson? Sometimes the innovation isn’t a better model, maybe the model gets less junk to read.
That takes us to TechRadar reporting on a familiar paradox in open-source cybersecurity. AI is making it easier to find and submit potential vulnerabilities. Unfortunately, it’s also making it easier to clog maintainers with inaccurate, duplicate, or poorly validated reports. TechRadar cites an expectation of around 66,000 unique vulnerability entries in 2026. And the challenge isn’t just finding bugs, it’s separating real exploitable issues from noise fast enough to do something about them. Think about the typical open-source maintainer. They may have a day job, a project used by thousands of orgs, and now an inbox full of machine-generated security reports. We’ve automated the creation of more work for a person who could be used to perform it.
But AI can also help triage reports, spot duplicates, prioritize severity, and suggest fixes. The critical part is validation. A confident-looking AI report is not the same as a confirmed vulnerability, and an AI-generated patch isn’t automatically a safe patch. The goal shouldn’t be the most findings, it should be more near promise fit. With less wasted engineering time.
And that brings us to a post by Fungus titled “Agentic Costs Are Unsustainable.” The premise is worth examining. Once an AI system starts actively planning, calling tools, scheduling, training, and passing work between models, the economics can change dramatically. A single chatbot response is one thing. An agent that makes dozens of calls to complete a task is something else entirely. Every extra step can add more tokens, tool costs, latency, and infrastructure usage. And when an agent gets stuck in a loop, you’re not just losing time, you’re paying for that loop.
The business equation is not whether agents can do impressive things, they can, it’s whether the value of the finished task exceeds the cost of running and supervising the system. That means measuring cost per outcome, successful outcome even, setting budgets and stop conditions, choosing smaller models when they’re sufficient, and keeping the human involved where mistakes are expensive. If an agent spends $20 doing a task that takes a person 5 minutes, we should probably stop calling that efficiency.
So here’s the thread connecting today’s stories. Markdown can reduce the information overhead we feed to AI. In security, AI can reduce human workload or multiply it if we don’t validate. And with agents, automation only pays off when the cost is getting to the answer. The cost of getting the answer makes sense. That’s the same principle 3 times. Useful output matters more than sheer volume of activity.
And that’s it for today on Open Channels News. I’m Robert Jacobi. Thanks for watching. And as always, be open and stay secure.
