Skip to main content

Posts

October 2025: AI updates from the past month

OpenAI announces agentic security researcher that can find and fix vulnerabilities OpenAI has released a private beta for a new AI agent called Aardvark that acts as a security researcher, finding vulnerabilities and applying fixes, at scale. “Software security is one of the most critical—and challenging—frontiers in technology. Each year, tens of thousands of new vulnerabilities are discovered across enterprise and open-source codebases. Defenders face the daunting tasks of finding and patching vulnerabilities before their adversaries do. At OpenAI, we are working to tip that balance in favor of defenders,” OpenAI wrote in a blog post . The agent continuously analyzes source code repositories to identify vulnerabilities, assess their exploitability, prioritize severity, and propose patches. Instead of using traditional analysis techniques like fuzzing of software composition analysis, Aardvark uses LLM-powered reasoning and tool-use. Cursor 2.0 enables eight agents to work in pa...
Post a Comment
Read more
Recent posts

OpenAI announces agentic security researcher that can find and fix vulnerabilities

OpenAI has released a private beta for a new AI agent called Aardvark that acts as a security researcher, finding vulnerabilities and applying fixes, at scale. “Software security is one of the most critical—and challenging—frontiers in technology. Each year, tens of thousands of new vulnerabilities are discovered across enterprise and open-source codebases. Defenders face the daunting tasks of finding and patching vulnerabilities before their adversaries do. At OpenAI, we are working to tip that balance in favor of defenders,” OpenAI wrote in a blog post . The agent continuously analyzes source code repositories to identify vulnerabilities, assess their exploitability, prioritize severity, and propose patches. Instead of using traditional analysis techniques like fuzzing of software composition analysis, Aardvark uses LLM-powered reasoning and tool-use. It is designed to work alongside developers and also integrates with existing workflows like GitHub and Codex so that it can provi...
Post a Comment
Read more

OpenAI completes restructuring, strikes new deal with Microsoft

OpenAI today announced that it has completed the restructuring of its business. When the company was founded in 2015, it was launched as a non-profit organization and that non-profit has controlled the for-profit arm of the business. Today’s restructuring turns the for-profit arm into a public benefit corporation called OpenAI PBC. The OpenAI Foundation—the new name for the non-profit—will still control the for-profit and hold a 26% equity stake in OpenAI PBC, which is currently valued at around $130 billion. Being a public benefit corporation differs from traditional corporate structures in that they are “required to advance its stated mission and consider the broader interests of all stakeholders, ensuring the company’s mission and commercial success advance together,” OpenAI’s website explains . According to Bret Taylor, chair of the OpenAI board of directors, the more OpenAI succeeds, the more the OpenAI Foundation’s equity will be worth, which it can use to fund philanthropic w...
Post a Comment
Read more

Gaining AI insights from spatial data

Spatial data – a record of physical or virtual data – is important to a variety of industries, yet a gap remains between collecting the raw data and gaining AI insights from it. I recently had the opportunity to speak with Damian Wylie, the head of products at spatial ETL, analytics and GeoAI company Wherobots , about the challenges of working with spatial data. This conversation has been edited for length and clarity. Q: What was the problem you saw with gleaning AI insights from spatial data? A: Let’s first start with what spatial data is, and then we can drill into some of the problems. So spatial data is a record of places, objects or activities, say, in a virtual or physical space. A virtual space could be something like a Metaverse or a game or an application. We’re going to spend most of our time today talking about the physical space. The physical space is anything tangible. This could represent things above our atmosphere, in space or in deep outer space, or could also be t...
Post a Comment
Read more

Cursor 2.0 enables eight agents to work in parallel without interfering with each other

The AI coding editor Cursor announced the launch of Cursor 2.0, the next iteration of the platform, featuring a new interface for working with multiple agents and its first ever coding model. The new multi-agent interface centers around agents instead of files. With this new interface, up to eight agents can work in parallel, using git worktrees and remote trees to prevent them from interfering with each other. It also allows developers to have multiple models attempt the same problem and see which one produces the best output. While this new interface is designed for agents, developers will still be able to open files or switch back to the classic IDE as needed. The new coding model, Composer, is four times faster than similar models, the company claims. It was designed for low-latency agentic coding tasks in Cursor, and it can complete most turns in less than 30 seconds. It was trained on a variety of tools, including codebase-wide semantic search, which makes it capable of under...
Post a Comment
Read more

Workato launches Enterprise MCP for SaaS platforms

Organizations are spending huge dollars on AI agents, but are finding that integrating the agents into all the systems the business needs to function is a very high hurdle. To help make SaaS platforms agent-ready, integration orchestration company Workato released Workato Enterprise MCP , which the company said in its announcement can “turn existing workflows, integrations, and APIs into rich, multi-step agent skills that any large-language-model (LLM)-based agent can call, including ChatGPT, Claude, Gemini, and Cursor.” Adam Seligman, chief technology officer at Workato, told SD Times that “the thing we keep coming back to over and over again is agents show a lot of promise, but to really work for business, they have to get access to business data. And they have to be able to do things inside your business, but do it in a way that you trust. And it’s really hard to get those two things right.” One of the reasons that organizations aren’t yet trusting their agents to do the right th...
Post a Comment
Read more

GitHub unveils Agent HQ, the next evolution of its platform that focuses on agent-based development

During its annual conference, GitHub Universe, GitHub shared its plans for Agent HQ, its vision for the future of the platform where AI agents are natively integrated across all of GitHub. As part of this Agent HQ initiative, over the next several months, paid GitHub Copilot users will gain direct access to popular coding agents from Anthropic, OpenAI, Google, Cognition, xAI, and more. Agent HQ brings with it several new capabilities to support this next evolution, the first of which is mission control, a central command center for assigning, steering, and tracking the work of multiple agents across GitHub, Copilot CLI, and VS Code. Mission control’s branch controls gives developers granular oversight over running checks for code created by the agents. Identity features will also be introduced to allow developers to manage agents like they would other coworkers and control which agent is building a task, manage access, and implement policies. Other mission control features include ...
Post a Comment
Read more

Eclipse Foundation launches ADL, an open language for defining agent behavior

The Eclipse Foundation today introduced the Agent Definition Language (ADL), an open language and visual toolkit for defining agent behavior. It was introduced as a part of the Eclipse Language Models Operating System (LMOS) project, an open source platform for building and running multi-agent systems. “Agentic AI is redefining enterprise software, yet until now there has been no open source alternatives to proprietary offerings,” said Mike Milinkovich, executive director of the Eclipse Foundation. “With Eclipse LMOS and ADL, we’re delivering a powerful, open platform that any organisation can use to build scalable, intelligent, and transparent agentic systems.” According to the Eclipse Foundation, ADL allows both business and engineering teams to collaborate on defining agent behavior in a maintainable and versionable way. It separates business logic from prompts, which makes it easier to build agents that can change. It is designed so that engineers set it up initially and then ...
Post a Comment
Read more

Chaos Engineering is non-negotiable in the AI era

We’ve all witnessed the AI boom over the past few years, but these seismic tech shifts don’t just materialize out of thin air. As companies rush to deploy AI models and AI-powered apps, we’re seeing a parallel surge in complexity. That growth is a threat to your system’s uptime and availability. It boils down to the sheer volume of interconnected components and dependencies. Each one introduces a new failure point that demands rigorous validation. This is exacerbated when, at the same time, AI is accelerating deployment velocities. This is why Chaos Engineering has never been more critical. And not as a sporadic check-the-box activity, but as a core, organization-wide discipline. Fault Injection via Chaos Engineering is the proven method to uncover failure modes lurking between services and apps. Integrate it into your testing regimen to plug those holes before they  trigger expensive incidents. Chaos Engineering Was Born in a Tech Explosion Those of us who’ve been around a wh...
Post a Comment
Read more

OpenAI completes restructuring, strikes new deal with Microsoft

OpenAI today announced that it has completed the restructuring of its business. When the company was founded in 2015, it was launched as a non-profit organization and that non-profit has controlled the for-profit arm of the business. Today’s restructuring turns the for-profit arm into a public benefit corporation called OpenAI PBC. The OpenAI Foundation—the new name for the non-profit—will still control the for-profit and hold a 26% equity stake in OpenAI PBC, which is currently valued at around $130 billion. Being a public benefit corporation differs from traditional corporate structures in that they are “required to advance its stated mission and consider the broader interests of all stakeholders, ensuring the company’s mission and commercial success advance together,” OpenAI’s website explains . According to Bret Taylor, chair of the OpenAI board of directors, the more OpenAI succeeds, the more the OpenAI Foundation’s equity will be worth, which it can use to fund philanthropic w...
Post a Comment
Read more

Forrester shares its predictions for how AI will continue to shape software development in 2026

As another year draws to a close, experts have begun looking ahead to how the technology landscape will evolve over the course of the next one. Forrester today released its 2026 predictions , with many of them relating to how software development will continue to be impacted by AI. Vibe coding took off in 2025, but next year, Forrester thinks it will evolve into vibe engineering, moving from just generating code to encompassing the full software development lifecycle. Today’s AI tools often produce problematic code, but Forrester believes these tools may improve enough in 2026 that they will be able to deliver engineering-grade outputs. The analysts also believe that there will be 20% fewer students enrolling in CS programs at universities, citing recent difficulties in new graduates finding jobs. They recommend CS professors update their courses to include AI material so that students will have the AI development skills employers are looking for. At the same time, it will take twi...
Post a Comment
Read more

Microsoft announces public preview for planning to improve how Copilot in Visual Studio handles complex tasks

Microsoft has announced a public preview for a new feature that aims to enable Copilot in Visual Studio to tackle more complex projects. With its new planning capability in Agent Mode, Copilot will research the codebase to break down big tasks into smaller and more manageable tasks, while also iterating on its plan as it works through the steps. “Planning makes Copilot more predictable and consistent by giving it a structured way to reason about your project. It builds on techniques from hierarchical and closed-loop planning research – enabling Copilot to plan at a high level, execute step-by-step, and adjust dynamically as it learns more about your codebase and issues encountered during implementation,” Rhea Patel, product manager at Microsoft, wrote in a blog post . In practice, when Copilot receives a prompt, it will decide if it should respond directly or switch into planning. Simple prompts will receive a quick answer, while multi-step prompts will trigger planning. Planning wi...
Post a Comment
Read more

GitKraken releases Insights to help companies measure ROI of AI

GitKraken, a software engineering intelligence company that specializes in improving the developer experience, announced the launch of GitKraken Insights to provide companies with better insights into AI’s impact on developer productivity. According to the company, while many engineering teams have adopted AI at this point, it is still a challenge to prove AI’s ROI. GitKraken also believes that traditional engineering metrics weren’t designed for the AI era. Matt Johnston, CEO of Gitkraken, told SD Times that despite the incremental investments in and perceived velocity gains from AI, they struggle to understand the impact. “I was talking to a VP of developer experience at a large Silicon Valley company, and he was basically saying,  ‘We’ve made investments of thousands of seats in Cursor and Copilot and Cloud, and we can’t really tell what’s being used… and how the heck do I measure this in a way that’s compelling to my business leaders.” GitKraken Insights brings together s...
Post a Comment
Read more

AWS updates its tool for measuring customer’s carbon footprints from using AWS services

AWS has announced updates to its Customer Carbon Footprint Tool (CCFT) , its solution that allows customers to track the carbon emissions their AWS usage causes. The CCFT uses the Greenhouse Gas Protocol ’s classification system that groups emissions into three categories. Scope 1 measures direct emissions, such as those from company facilities or vehicles, while Scope 2 and 3 measure indirect emissions. Scope 2 focuses on upstream emissions produced indirectly, such as fuel and energy use or waste generated during operations, while Scope 3 measures indirect downstream emissions, such as those from transportation and distribution of acquired products or end-of-life treatment of those products. Previously, the CCFT measured Scope 1 and 2 emissions, but now the company is also incorporating Scope 3 emissions, as well as adding another category to Scope 1 emissions. CCFT now tracks refrigerants and natural gas in its Scope 1 emissions, adding to the existing Scope 1 emissions coming fr...
Post a Comment
Read more

This week in AI updates: mabl Agentic Testing Teammate, Couchbase 8.0, and more (October 24, 2025)

Mabl announces updates to Agentic Testing Teammate The Agentic Testing Teammate works alongside human testers to make the process more efficient. New updates include AI vectorizations and test semantic search, improvements to test coverage, and enhancements to the MCP Server that enable testers to do a number of tasks directly within their IDE, including Test Impact Analysis, intelligent test creation, and failure recommendations. “This new work is built on the idea that an agent can become an integral part of your testing team,” said Dan Belcher, co-founder of mabl. “Unlike scripting frameworks and general-purpose large language models, mabl builds deep knowledge about your application over time and uses that knowledge to make it–and your team–more effective.” Couchbase 8.0 adds three new vector indexing and retrieval capabilities These new capabilities are designed to support diverse vector workloads that facilitate real-time AI applications. Hyperscale Vector Index is based on ...
Post a Comment
Read more