If you find our site valuable and helpful, consider supporting us or sponsoring one of our curations.
Source/Link | Description |
|---|---|
InfoWorld - What happens when engineering teams reorganize around AI agents | But what made Camp AI’s “Agents at Work” event (hosted by Auth0) stand out was its showcase of companies that are in various stages of reorganizing their engineering processes around AI agents. Browserbase, Mastra, Fireworks AI, Drata, Mya, MindFort, and Corridor are all part of the vendor ecosystem trying to enable secure and performant agentic AI, but the most revelatory stories were their own successes and the challenges they faced restructuring their engineering orgs for agents. |
Computerworld - US government agency to safety test frontier AI models before release | According to a release from CAISI, which is part of the department’s National Institute of Standards and Technology (NIST), it will “conduct pre-deployment evaluations and targeted research to better assess frontier AI capabilities and advance the state of AI security.” |
InfoWorld - Three skills that matter when AI handles the coding | But that shift creates a new challenge nobody is talking about enough. As AI takes on the heavy lifting, the skills that matter most are moving upstream: how to provide the right context for a prompt, how to evaluate what the model produces, and how to understand a problem deeply enough that you can’t be fooled by a confident but wrong answer. |
Agentic AI promises faster coding, but hidden risks in testing, security, and maintenance could derail projects unless developers rethink how they manage, validate, and supervise machine-generated software at scale | |
InfoWorld - What happens when engineering teams reorganize around AI agents | But what made Camp AI’s “Agents at Work” event (hosted by Auth0) stand out was its showcase of companies that are in various stages of reorganizing their engineering processes around AI agents. Browserbase, Mastra, Fireworks AI, Drata, Mya, MindFort, and Corridor are all part of the vendor ecosystem trying to enable secure and performant agentic AI, but the most revelatory stories were their own successes and the challenges they faced restructuring their engineering orgs for agents. |
The Algorithmic Bridge Substack - How the AI Industry Runs on Its Own Money | Half of the revenue backlog reported by four of the largest companies on earth comes from two startups fully dependent on the funding and patronage of those four companies. The circularity of the AI industry strikes again. |
InfoWorld - Building AI apps and agents with Microsoft Foundry | Microsoft Foundry helps application developers to build and deploy agents, which may use models and tools. It also helps machine learning (ML) engineers and data scientists to fine-tune models, run evaluations, and manage model deployments. Finally, it helps IT administrators and platform engineers to govern AI resources, enforce policies, and manage access across teams. It isn’t quite a floor wax and a dessert topping, but it does try to serve three distinct audiences. |
ZDNet - Your Claude agents can 'dream' now - how Anthropic's new feature works | On Wednesday, during its Code with Claude event, Anthropic updated Managed Agents with a new feature called "dreaming," which lets agents "self-improve" by reviewing past sessions for patterns, according to Anthropic. Building on an existing memory capability, the feature schedules time for agents to reflect on and learn from their past interactions. Once dreaming is on, it can either automatically update your agents' memories to shape future behavior or you can select which incoming changes to approve. |
Brox, a predictive human intelligence startup, recently announced a strategic funding round following a year where they reported 10X revenue growth. Their proposition is as ambitious as it is technical: the creation of a "parallel universe" populated by 60,000 digital twins of real, living human beings and their entire demographic profiles and consumer preferences, allowing enterprises to run unlimited experiments in hours rather than months. | |
SiliconANGLE - Monday.com relaunches as an AI work platform with native agents | The company’s new AI agents ship inside the platform and are available to every customer with no setup required. The agents can draft marketing campaigns, qualify sales leads, triage support tickets, generate reports, run project workflows and handle budget approvals. The agents operate under human supervision and within monday.com’s existing security, permissions and governance controls. |
The company claims its first model, SubQ 1M-Preview, is the first LLM built on a fully subquadratic architecture — one where compute grows linearly with context length. If that claim holds, it would be a genuine inflection point in how AI systems scale. At 12 million tokens, the company says, its architecture reduces attention compute by almost 1,000 times compared to other frontier models — a figure that, if validated independently, would dwarf the efficiency gains of any existing approach. | |
OpenAI ships its GPT models in tiers. Two weeks ago, it announced the launch of GPT-5.5 Thinking and Pro, which are designed for slower, more analytical responses and for memory-intensive tasks, respectively. However, it’s the Instant version that serves as ChatGPT’s default, and the one most users will interact with, unless they go to the settings to adjust things. | |
CLI-Anything generates SKILL.md files, the same instruction-layer artifacts that Snyk’s ToxicSkills research found laced with 76 confirmed malicious payloads across ClawHub and skills.sh in February 2026. A poisoned skill definition does not trigger a CVE and never appears in a software bill of materials (SBOM). No mainstream security scanner has a detection category for malicious instructions embedded in agent skill definitions, because the category simply did not exist eighteen months ago. | |
SiliconANGLE - ServiceNow bids to become the control tower for enterprise AI | The announcements center on enhancements to the company’s AI Control Tower (pictured), a governance hub that allows organizations to monitor, manage and secure AI models and agentic workflows. The company also introduced new autonomous security capabilities and an expanded “Autonomous Workforce” of AI agents designed to execute specific business processes. |
Agents aren't one-off builds. "The problem is you end up with dozens of agents with no shared context model, no consistent governance, and no reusable patterns," Sanyal said. "A proper management platform gives you composable primitives, multi-tenant isolation, model routing across LLM providers, and observability into what agents are actually doing." | |
While the current administration has appeared to embrace a laissez-faire approach to powerful AI systems and their development, safety concerns have mounted since the partial release of Anthropic PBC’s Claude Mythos model. “Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” said CAISI Director Chris Fall in a statement. “These expanded industry collaborations help us scale our work in the public interest at a critical moment.” | |
ZDNet - Google's AI Overviews will show you advice from other people now | Now, AI Overviews will show you advice from other people. Appearing in a section called Expert Advice, these comments will appear as brief remarks from people via discussion forums, social media, and other online sources. Each comment will show you the name of the person or the forum and include a link that will take you to the full discussion. From there, you can read more comments about your topic and even join the discussion if you'd like. |
SmartR AI Substack - Fractile and the Art of Analog AI – A Skeptical Engineer’s Deep Dive | At its heart, the company is designing a memristor crossbar – a grid of programmable memory elements that can perform matrix multiplication in a single step, directly in memory. The input voltage is converted to a current via Ohm’s Law (I = V × G), and Kirchhoff’s Current Law physically sums the currents along each column to produce a dot product. |
The retrieval-augmented generation (RAG)-to-vector database pipeline doesn't cut it anymore; agentic AI requires a different approach that incorporates context. VentureBeat's Q1 2026 Pulse survey underscores this trend: Every standalone vector database is losing adoption share, while hybrid retrieval intent has tripled to 33.3%, the fastest-growing strategic position in the dataset. | |
ZDNet - What you'll pay for AI agents will be wildly variable and unpredictable | Among all the challenges of implementing agentic artificial intelligence, the least-understood issue is cost. The providers of AI, such as OpenAI, Google, and Anthropic, have price lists, but none of those listed prices tell users what the final bill will be to actually solve a problem. |
Computerworld - Relying on LLMs is nearly impossible when AI vendors keep changing things | Over the years, enterprise IT execs have gotten frighteningly comfortable having little control or visibility over mission-critical apps, from SaaS to cloud and even cybersecurity. But generative AI (genAI) and agentic systems are taking that problem to a new extreme, with vendors able to dumb down a system IT is paying billions for without so much as a postcard. |
VentureBeat - The AI scaffolding layer is collapsing. LlamaIndex's CEO explains what survives. | Further, coding agents excel at writing code, meaning devs don’t need to rely on extensive libraries. In fact, about 95% of LlamaIndex code is generated by AI. “Engineers are not actually writing real code,” Liu said. “They're all typing in natural language.” This means the layers between programmers and non-programmers is collapsing, because “the new programming language is essentially English.” |
SiliconANGLE - Pentagon inks AI procurement deals with seven companies, leaves out Anthropic | The AI products covered under the newly issued contracts will be used in Impact Level 6 and Impact Level 7 environments. Those are Defense Department systems that can be used to store classified information. According to Nextgov/FWC, the goal of the AI initiative is to ease data synthesis and streamline decision-making workflows. |
VentureBeat - 200,000 MCP servers expose a command execution flaw that Anthropic calls a feature | Anthropic created the Model Context Protocol as the open standard for AI agent-to-tool communication. OpenAI adopted it in March 2025. Google DeepMind followed. Anthropic donated MCP to the Linux Foundation in December 2025. Downloads crossed 150 million. Then four researchers at OX Security found an architectural problem that affects all of them |
What’s becoming more clear is that hyperscalers are the true landowners of AI. They are the custodians and the ones that benefit directly from the demand for compute in more Cloud computing and digital Advertising revenue. This week’s Earnings bonanza of BigTech (4 of the Mag 7 reported after the close yesterday) really displayed this 4 years later (ChatGPT debuted at the end of 2022) in more concentrated terms than ever before. A new world of technological domination is coming into view. | |
SiliconANGLE - OpenMetadata maker Collate launches AI Analytics for chat-driven dashboards | Collate, the company behind the open-source OpenMetadata project, is pitching the product as allowing analysts to work without prior knowledge of where data lives or how metrics are defined inside an organization. Analysts type a question and the tool identifies the relevant data, generates the query and visualizes the result. Charts can be saved, shared and assembled into dashboards used across the enterprise. |
VentureBeat - Netomi raises $110 million as Accenture and Adobe bet on AI for customer service | The answer, in Mehta's view, is not to make downstream customer service faster with AI. It is to prevent the service ticket from being created in the first place. That logic sits behind almost every strategic decision the company has made — including the Adobe partnership. |
Computerworld - AI venture funding to shoot up this year as bubble looms | Gold sees clear signs of an AI bubble, starting with the murky path to profitability for AI firms. “Does AI have the potential to generate lots of revenues? Yes, but with the current spend rate on infrastructure, it’s hard to see how [AI vendors] can be revenue-positive in the short term (two to three years),” he said. |
Humanity Redefined Substack - Walled Artificial Intelligence - by Conrad Gray | This is an important distinction to keep in mind. You do not need access to a model to benefit from what it produces. If an AI plays an instrumental role in discovering a new therapy that cures your cancer, you benefit, even though you might never have been able to access the model. If it identifies a critical vulnerability in the software your bank runs on and that vulnerability gets patched before anyone exploits it, you benefit. But the model stays behind the wall. |
VentureBeat - How to build custom reasoning agents with a fraction of the compute | The standard method for training reasoning models is Reinforcement Learning with Verifiable Rewards (RLVR). In this paradigm, the model learns through trial and error, guided by a final outcome from its environment. An automated verifier checks if the model’s answer is right or wrong, providing a binary reward, such as a 0 or 1. |
SiliconANGLE - Salesforce introduces Agentforce Operations to automate outdated back-office tasks | Parulekar said Agentforce Operations is “sort of the relaunch of that for more industries than they had in the past.” She added that Regrello’s work helped Salesforce think through the deeper pain points of process automation: how to let business users upload a process and get immediate optimization, while still preserving determinism and human oversight where it matters. |
One of the more dangerous assumptions in the current AI market is that broad adoption means meaningful adoption. It does not. Much of what enterprises call AI transformation is, in fact, AI experimentation focused at the edge of the business, in systems and workflows that support employees but are not central to how the enterprise actually operates. These include calendaring, scheduling, meeting summaries, employee communications, customer messaging, document generation, internal assistants, and similar productivity-oriented use cases. | |
The Leverage Point Substack - Most AI Training Is a Waste of Money - by Kyle Behrend | Not because training isn’t important. It is. But because training without strategy is just an event. People show up, they nod along, they learn a few things. Three weeks later, nothing has changed. Leadership wonders why adoption isn’t sticking. And somebody starts googling the next workshop. |
While it is correct to say Apple has had setbacks in AI development, it has also had successes. Was it ready for the generative AI (genAI) juggernaut? Probably not, nor has it successfully developed its own response in-house. Is Apple’s platform ready for AI? Indisputably, with the power and performance across all its hardware products to run AI on the edge, in the cloud, and as-a-service. Right now, Apple doesn’t offer the world’s best AI services, but does offer the world’s best platform on which to run them. | |
VentureBeat - Context decay, orchestration drift, and the rise of silent failures in AI systems | We have spent the last two years getting very good at evaluating models: benchmarks, accuracy scores, red-team exercises, retrieval quality tests. But in production, the model is rarely where the system breaks. It breaks in the infrastructure layer, the data pipelines feeding it, the orchestration logic wrapping it, the retrieval systems grounding it, the downstream workflows trusting its output. That layer is still being monitored with tools designed for a different kind of software. |
AI Supremacy Substack - Summary of the AI Index Report 2026, Part II | It’s not just an AI paradox, it’s a legitimate AI risk that I’ve not seen Think Tanks and AI policy academics report on much. Bottom up consensus on AI is starting to form, and it’s contrary to what a lot of Executives, elites and BigTech lobbying has been suggesting. It’s this National divide in the U.S. with regards to the benefits and risks of AI that most concerns me. |
VentureBeat - Monitoring LLM behavior: Drift, retries, and refusal patterns | While using one non-deterministic system to evaluate another seems counterintuitive, it is an exceptionally powerful architectural pattern for use cases requiring nuance. It is virtually impossible to write a reliable regex to verify if a response is "actionable" or "polite." While human reviewers excel at this nuance, they cannot scale to evaluate tens of thousands of CI/CD test cases. Thus, the LLM-as-a-Judge becomes the scalable proxy for human discernment. |
SiliconANGLE - The third leg of AI’s infrastructure race isn’t silicon or power. It’s capital | The next era of AI isn’t being defined by who can build the fastest chip or secure the cheapest power. It’s being defined by who can bridge the massive gap between raw technical demand and institutional bankability. Enter Argentum AI, a company that I interviewed a few months ago has moved from relative obscurity to aggregating a staggering $50 billion in demand interest, representing more than 400,000 graphics processing units. |
2nd Order Thinkers Substack - I Read All 414 Pages of Stanford's AI Index. The Summary Misleads. | Stanford’s report put together frontier models against the human baseline on eleven benchmarks — image recognition, language understanding, reading comprehension, competition maths, PhD-level science, multimodal reasoning, visual reasoning, and a couple more — this year. |
AI, academia, and the Future Substack - The economics of AI in spring 2026 - by Bryan Alexander | The subscription model seems to be working well for Anthropic. A Bloomberg report found the Claude-maker’s numbers increasing, especially after the fight with the Pentagon, and projected in March the company would earn $20 billion annually as a result, what with 1,000 businesses paying at least $1 million each for Claude. In April Anthropic estimated it would take in $30 billion. |
The product, which launches as part of Mistral's Studio platform, is the company's clearest articulation yet of a thesis that is quietly reshaping the enterprise AI market: that the bottleneck for organizations adopting AI is no longer the model itself, but the infrastructure required to run it reliably at scale. | |
I have been encouraging you to think about AI not as a single thing, but as a set of three interlinked concepts. You need to consider models, like Opus 4.7, Gemini 3.1, or (now) GPT-5.5. You also want to pay attention to apps, which are the products you actually use to talk to a model, and which let models do real work for you. The most common app is the website for each of these models: chatgpt.com, claude.ai, gemini.google.com. But, increasingly, desktop applications like Claude Code, Claude Cowork, and OpenAI Codex are becoming the most useful apps for AI | |
Cognition is best known for its AI programming assistant Devin, which burst onto the scene in a big way in March 2024, helping to accelerate the “vibe coding” trend. At the time, the startup claimed Devin was the world’s first “fully autonomous AI software engineer,” rather than a simple coding tool. It’s designed to plan and execute complex engineering tasks by automating the code creation, debugging and even the deployment process. | |
Computerworld - The agentic AI frenzy increases as more vendors stake their claims | Right now, he added, “many agents are being treated as simply a more advanced front end. These newer options provide the ability for an agent to do things like spin up a dedicated container, and they can support semi-autonomous and, in some cases, autonomous operations. These two announcements are more infrastructure-related, whereas OpenAI is more about agent building.” |
ZDNet - I put GPT-5.5 through a 10-round test: It scored 93/100, losing points only for exuberance | Since GPT-5.5 is only available in paid tiers (Plus and above), I used ChatGPT Plus for my tests. Right now, my Plus account only shows GPT-5.5 available for the Thinking effort level in both Standard and Extended. I picked Standard Thinking. That's the effort I used for these tests. |
Computerworld - Tim Cook’s legacy: a successful CEO who stumbled over AI | The company did not believe in AI until ChatGPT emerged to take the tech industry by storm, according to a 2025 Bloomberg news report, and Cook did not provide the resources needed in-house to develop an AI-powered Siri. Apple was forced in early 2025 to look at partnering with Anthropic and OpenAI to push Siri into the AI era; in finally settled on working with Google’s Gemini. |
Called "Workspace Agents," OpenAI's new offering essentially allows users on its ChatGPT Business ($20 per user per month) and variably priced Enterprise, Edu and Teachers subscription plans to design or select from pre-existing agent templates that can take on work tasks across third-party apps and data sources including Slack, Google Drive, Microsoft apps, Salesforce, Notion, Atlassian Rovo, and other popular enterprise applications. | |
InfoWorld - Google’s Gemma 4 shines on local systems – both big and small | What’s striking about Gemma 4 is that even at the higher end of its size range, it’s still decently performant on personal hardware. Google claims this is due to innovations in the architecture of the model, but the proof is in the trying. Gemma 4 is quite responsive. |
SiliconANGLE - OpenAI subscribers get new ‘workspace agents’ to automate complex tasks across teams | They’re meant to be the next evolution of the company’s “GPTs,” which are specialized, no-code versions of ChatGPT tailored by their users to perform specific tasks, hobbies and workflows. Powered by Codex, they can take on many of the tasks people already do at work — from preparing reports to writing code to responding to messages, the company said: “They run in the cloud, so they can keep working even when you’re not. They’re also designed to be shared within an organization, so teams can build an agent once, use it together in ChatGPT or Slack, and improve it over time.” |
Computerworld - Google Chat becomes an agent interface for Workspace | The Ask Gemini in Chat feature — generally available now — also allows Workspace users to interact with third-party tools such as Asana, Jira, and Salesforce, Google said in a blog post. Ask Gemini in Chat provides a “daily briefing” that highlights action items, unread message threads, and tasks to complete. It can also act on behalf of a user, generating documents and slides, scheduling meetings, and retrieving files, for instance. |
AI Supremacy Substack - Why Cursor is the Enterprise AI Darkhorse of Generative AI | Cursor (the flagship product of the startup Anysphere) is building one of the most viable “vibe-working” platforms in the world. This is the sort of interface and tech that can enable one high performing employee to do the job of what would have taken entire teams just a few years ago to accomplish. |
SiliconANGLE - With Workspace Intelligence, Google’s productivity suite gets more AI smarts | The main update is a new system called Workspace Intelligence, announced today at Google Cloud Next 2026, that helps transform simple applications into autonomous and collaborative AI agents. In a blog post, Google Workspace Vice President of Product Yulie Kwon Kim said the company is no longer content with AI serving as a passive assistant that just sits and waits for a prompt. Instead, it’s pushing the idea of “agentic work,” where AI understands the context behind the projects someone is working on, and possesses the organizational knowledge to start working on them autonomously. |
AI Policy Perspectives Substack - Q&A with Ethan Mollick - by Tom Rachman | But with agentic systems that could do hours of work on their own, now it’s a design choice. When do we want humans-in-the-loop? When is human verification valuable? When is human verification morally required? When is it legally required? What kind of interventions move the system forward? I feel there has been a complete lack of deep understanding about these topics. |
InfoWorld - GitHub pauses new Copilot sign-ups as agentic AI strains infrastructure | “As Copilot’s agentic capabilities have expanded rapidly, agents are doing more work, and more customers are hitting usage limits designed to maintain service reliability,” GitHub said in a blog post. “Without further action, service quality degrades for everyone.” |
AI Supremacy Substack - Summary of the AI Index Report 2026 - by Michael Spencer | The jagged frontier speaks of a quickly developing industry that is a great lens on the State of AI over the past year 2025-2026 and the big picture. While a lot of the AI reports I read are from Venture Capital funds whose job it is to make AI look amazing, with an academic institution we can hope to find a less biased snapshot. I am not sure if that turns out to be true however. The AI Index that was established in 2019 as an independent project claiming to provide unbiased, rigorous data on the progress of AI. |
The AI Memo Substack - The End of the Smartest Person in the Room | In this issue, we unpack three shifts that are already redefining executive leadership: why ego is becoming a liability, how orchestration is replacing hierarchy, and why hands-on co-creation with AI is no longer optional. This is already happening inside organizations. If you’re leading teams, making decisions, or shaping strategy, these are quickly becoming the expectations of the day. |
Computerworld - Adobe builds an ‘agentic content supply chain’ for the AI era | The agent automates actions across workflows such as planning, execution, review, and approval. This can help teams build projects, speed up reviews, and pull out insights on demand, without the need for manual reporting. Teams can also involve AI agents in project planning and assign them tasks, such as resolving simple issues or performing reviews based on specific instructions and context. |
Beyond the AI Hype by WAIE+ Substack - Scaremonger, hypester, or responsible steward of AI? | Claude Mythos2 Preview is a general-purpose, unreleased frontier model that reveals a stark fact: AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities. |
The first wave of corporate AI training focused heavily on prompt engineering and basic familiarity with generative AI tools. That approach made sense early on, when employees needed help understanding the technology. But many organizations are discovering those skills have a short half-life. | |
InformationWeek - Meta's new 'AI Zuckerberg' is a mirror for every C-suite | Meta is building an AI version of Mark Zuckerberg, according to a report from the Financial Times earlier this week. The goal is for the digital proxy to interact with employees, field questions and simulate the executive presence of one of the most recognizable technology CEOs in the world. The immediate reaction -- somewhere between fascination and eye roll -- is understandable. But executives would be wise not to dismiss the announcement altogether. |
Computerworld - Google’s new AI app is a glimpse of the future | AI Edge Eloquent sherlocks Wispr Flow and Willow, which each cost $15 per month. It also sherlocks SuperWhisper, priced at $85 per year. (In Silicon Valley parlance, “sherlocking” is when a major company copies a major feature of a competitor’s product, thereby rendering the competitor’s product obsolete.) |
Salesforce on Wednesday unveiled the most ambitious architectural transformation in its 27-year history, introducing "Headless 360" — a sweeping initiative that exposes every capability in its platform as an API, MCP tool, or CLI command so AI agents can operate the entire system without ever opening a browser. | |
SiliconANGLE - AI Mode in Chrome adds split-screen view to enhance the web search experience | For instance, if someone is looking to buy a new coffee maker and wants some very specific features, they can tell AI Mode exactly what they’re looking for and it will pull up a list of suitable products. If the user clicks on one of the options, they’ll be able to browse that website with the AI Mode tab right there alongside it. They can then ask more pointed questions, such as “how easy is this to clean?” and AI Mode will scan both the open page and the wider web to generate an informed answer. |
Computerworld - Meta moves fast toward a world where AI builds the software | Meta Platforms is reportedly pulling top software engineers from across the company into a newly created AI unit on a mandatory basis, with the stated goal of eventually having autonomous agents perform the bulk of the work of building, testing, and shipping its products, and human engineers serving only to monitor them. |
VentureBeat - Claude, OpenClaw and the new reality: AI agents are here — and so is the chaos | The key to making these tools more impactful is giving them more power, but that increases the risk of misuse. Users must trust providers like Anthorpic and Google to ensure that agent prompts will not cause harm, leak data, or provide unfair (illegal) advantage to certain vendors. OpenClaw is open-source, which complicates things, as there is no central governing authority. |
Computerworld - Google’s new AI app is a glimpse of the future | Cloud computing is necessary for chatbots like ChatGPT, Perplexity, Claude and Gemini because all-purpose AI models require hundreds of billions of parameters, massive amounts of RAM, and huge amounts of electricity to be ready to do anything and everything very quickly. Forcing these workloads onto a mobile device fundamentally caps the intelligence and capability of general-purpose AI. But breaking down individual tasks (like transcription) doesn’t require massive data centers. |
Now, today, Meta is showing us the fruits of that effort: Muse Spark, a new proprietary model that Wang says (posting on rival social network X, used more often by the machine learning community) is "the most powerful model that meta has released," and has "support for tool-use, visual chain of thought, & multi-agent orchestration." He also says it will be the start of a new Muse family of models, raising questions about what will become of Meta's popular lineup and ongoing development of the Llama family. | |
InfoWorld - Final training of AI models is a fraction of their total cost | This has been a concern of US AI companies for some time. Google has already expressed concerns about intellectual property theft. And Anthropic has fingered MiniMax as a company that has sought to extract Claude’s capabilities to enhance its own offerings. It’s clear that any business looking to develop AI models is going to be committing to spend huge sums of money: The training is just a small part of it. |
SiliconANGLE - Microsoft accelerates agentic automation with Copilot Cowork for complex workflows | Spataro said users simply tell Copilot Cowork what they’re trying to accomplish, and it will then go ahead and create a plan before immediately carrying out the necessary tasks to achieve that goal, reasoning across various Microsoft 365 applications and files. Human oversight is still present, though. As it’s working, humans will be able to monitor the agent’s progress, and step in to “steer” it in the right direction should it go off track, Spataro said. |
VentureBeat - When product managers ship code: AI just broke the software org chart | For a meaningful class of tasks, it became faster to just build the thing than to describe what you wanted and wait for someone else to build it. Think about that for a second. Every modern software organization is structured around the assumption that implementation is the expensive part. When that assumption breaks, the org has to change with it. |
Forbes - CNCF's Dapr Agents Tackles The Problem Most AI Frameworks Ignore | The Cloud Native Computing Foundation announced the general availability of Dapr Agents v1.0 at KubeCon Europe in Amsterdam this week, releasing a Python framework that prioritizes keeping AI agents alive through crashes and failures over making them smarter. Zeiss Vision Care validated the approach in a KubeCon keynote, showing how the framework powers a durable document extraction pipeline processing enterprise optical data at scale. |
SiliconANGLE - Anthropic to launch new ‘Claude Mythos’ model with advanced reasoning features | Mythos came to light after the company accidentally left a CMS folder with 3,000 assets publicly accessible. The repository contained a draft version of a launch blog post for Claude Mythos. According to Fortune, the document indicates that the model will be pricier than the company’s existing algorithms. |
VentureBeat - When AI turns software development inside-out: 170% throughput at 80% headcount | Qualitatively, looking at the business value, I actually see even higher uplift. One reason is that, as we started last year, our quality assurance (QA) team couldn’t keep up with our engineers' velocity. As the company leader, I wasn’t happy with the quality of some of our early releases. As we progressed through the year, and tooled our AI workflows to include writing unit and end-to-end tests, our coverage improved, the number of bugs dropped, users became fans, and the business value of engineering work multiplied. |
SiliconANGLE - OpenAI introduces plugins for its Codex programming assistant | Plugins can also integrate Codex with external services using MCP servers. Developers may upload configuration files to customize how those MCP servers work. For example, an engineer could connect Codex to an MCP-powered development environment and specify what middleware should be pre-installed in the sandbox. |
SiliconANGLE - Oracle’s new AI bet: Make the AI database the center of agentic workloads | At its latest showcase during the Oracle AI World Tour London 2026, the company made a calculated move in the agentic AI market. Instead of competing head-on in the model wars, Oracle is positioning the database as the center of gravity for enterprise agentic AI, effectively arguing that the future of AI won’t be determined by agents alone, but by where and how they interact with data. |
Following the ruling, Anthropic issued a statement saying, “We’re grateful to the court for moving swiftly, and pleased they agree Anthropic is likely to succeed on the merits. While this case was necessary to protect Anthropic, our customers, and our partners, our focus remains on working productively with the government to ensure all Americans benefit from safe, reliable AI.” | |
The capability spread hasn’t gone to zero. Frontier models still lead on the hardest tasks, such as complex agentic coding, multi-step tool chaining, and long-horizon workflows where reliability matters as much as raw intelligence. But the list of tasks where they meaningfully lead is getting shorter every quarter. And that distinction, between leading on benchmarks and leading on tasks people pay for, is everything | |
VentureBeat - How xMemory cuts token costs and context bloat in AI agents | Experiments show that xMemory improves answer quality and long-range reasoning across various LLMs while cutting inference costs. According to the researchers, it drops token usage from over 9,000 to roughly 4,700 tokens per query compared to existing systems on some tasks. |
SiliconANGLE - Bland launches Norm to help teams build production-ready voice agents in minutes | The system would then go to work and automatically generate the prompt, persona, agent, pathways, validation conditions, extraction rules and API integrations. The engineering team can then thoroughly test the agent before deploying it live. |
Computerworld - Apple’s AI endgame: Why waiting for Siri could make it a winner | Bloomberg tells us Apple has a big plan for iOS 27 with a massive Siri revamp, turning it into a full-scale chatbot like ChatGPT or Gemini. The update might extend to Siri gaining its own dedicated Siri app, deeper system integration, and potentially replacing or integrating Spotlight search. The idea is that Siri will be as good as any other chatbot you use, but will also be equipped with information personal to you — only on the device, private, and secure (insofar as there is any privacy and security anymore). |
Forbes - What Sora’s Rise And Sudden Fall Means For OpenAI, Disney And AI Video | That sudden ending leaves two contrasting stories sitting side by side. One sees OpenAI reallocating resources toward coding, enterprise products, world simulation and robotics, where the strategic and financial upside is clearer. The other sees a company that overreached, launched a video product before it had a stable commercial and ethical footing, then pulled back once the costs, controversies and distractions became too difficult to justify. Reuters has reported that OpenAI is refocusing around coding and business users, while Axios says the Sora research team will continue “to focus on world simulation research to advance robotics that will help people solve real-world, physical tasks.” |
The update, available immediately as a research preview for paying subscribers, transforms Claude from a conversational assistant into something closer to a remote digital operator. It arrives inside both Claude Cowork, the company's agentic productivity tool, and Claude Code, its developer-focused command-line agent. Anthropic is also extending Dispatch — a feature introduced last week that lets users assign Claude tasks from a mobile phone — into Claude Code for the first time, creating an end-to-end pipeline where a user can issue instructions from anywhere and return to a finished deliverable. | |
SiliconANGLE - OpenAI says it’s pulling the plug on Sora, its generative AI video creation tool | Co-founder and Chief Executive Sam Altman revealed the decision today, telling staff in an email that the company will wind down all products that use its video models. It’s dropping both the consumer app and also a version for developers. ChatGPT will no longer support video functions either. In a social media post, OpenAI thanked Sora’s users, saying it understands that the news will be “disappointing.” |
InfoWorld - OpenAI’s desktop superapp: The end of ChatGPT as we know it? | “We realized we were spreading our efforts across too many apps and stacks, and that we need to simplify our efforts,” the Journal reported that day, citing Simo’s address to the employees. “That fragmentation has been slowing us down and making it harder to hit the quality bar we want.” At the same meeting, Simo outlined the commercial imperative plainly: “Our opportunity now is to take those 900 million users and turn them into high-compute users. We’ll do that by transforming ChatGPT into a productivity tool.” |
Many others similarly started the interview talking about productivity, but after Anthropic Interviewer asked about their underlying hope behind it—what realizing this vision would enable for them—other priorities surfaced. It wasn’t about doing better work, but increasing their quality of life outside of it. Using AI to automate e-mails became, in actuality, a desire to spend more time with family. | |
The Economy of Algorithms Substack - Stop Asking AI for Answers - by Marek Kowalkiewicz | Ninety agents reported that volume grew by about 6% per year. Sixty reported that volume declined by about 5% per year. Within each group, the agents agreed almost perfectly, within 0.25% and 0.11%, respectively. Precise. Consistent. And completely opposing one another. In a consulting firm, this would be two confident partners arguing at a board meeting. |
And if that’s where the world is headed, then speech and debate may be the single best training ground for the next generation of founders. Not because it teaches you to talk fast (though it does), but because it builds the cognitive architecture that entrepreneurship demands. Pair that foundation with fluency in multi-agent AI systems, and you have something genuinely powerful: a person who can think, persuade, adapt, and orchestrate — whether the team around them is human, artificial, or both. | |
Cyborgs Writing Substack - The Techne Behind Agent Skills - by Lance Cummings | Task information is exactly right for execution steps. But a skill is also a description of what the output is supposed to be, a set of behavioral constraints, a map of how this task connects to the larger workflow, and the metadata that triggers the skill in the first place. |
FullStack HR Substack - "Treat AI as if it were human because, in many ways, it behaves like one." | His concept of the “Never Normal,” that we should stop waiting for things to calm down because they never will, landed with particular force this year. He framed the current moment through four lenses: hardcore geopolitics, extreme capitalism, zero latency, and talent singularity. His conclusion was clear. AI-first does not mean people-last, but the waves of disruption are only going to get higher. |
The Augmented Educator Substack - Did the AI Bubble Burst? - by Michael G Wagner | The system initiated what engineers call context compaction, a lossy compression process that summarizes and discards tokens the algorithm deems non-essential. In this case, the foundational safety constraint — “don’t action until I tell you to” — was among the tokens pruned. Stripped of its guardrails, the agent defaulted to its primary objective of inbox optimization and launched what observers described as a “speedrun,” bulk-trashing and archiving hundreds of important personal emails across multiple accounts. |
AI Supremacy Substack - Cursor's Wild Trajectory to being a Vibe Working Leader | Cursor has demonstrated unprecedented scaling in the AI dev tools space, outpacing competitors like GitHub Copilot or Devin. I believe Cursor’s Enterprise AI suite will become a peer of that of Anthropic. In short, I don’t see them as just an AI coding startup any longer. |
Forbes - AGI Is Infeasible. Instead, Pursue Superhuman Adaptable Intelligence | SAI establishes a new goal for the industry. Instead of trying to build a singular, all-capable solution that can do everything a human can do, which is the goal represented by AGI, this framework champions a return to specialized, narrow AI. These researchers advocate leveraging massive amounts of data through self-supervised learning, as for example LLMs do, but they also advise that such tech then be adapted to solve specific problems. |
Computerworld - Amazon finds out AI programming isn’t all it’s cracked up to be | Amazon might yet regret that move. Multiple Amazon Web Services (AWS) and Amazon retail outages have prompted an internal crackdown on how generative AI (genAI) is allowed to touch production code. What’s’ that line about the horse and the barn door? |
Forbes - What If AI Isn’t A Bubble, But It Still Crashes The Economy? | Rather than a sudden, worldwide crash into economic depression, the scenario proposes a more gradual erosion of quality of life, spending power, access to opportunity and political freedom, as wealth and power are increasingly concentrated in the hands of the corporations that control AI. |
The Algorithmic Bridge Substack - How AI Will Erase Entire Industries Without Automating Them | Here’s an example: A few days ago, Andrej Karpathy described an experiment where he left a swarm of AI agents running for two days on his nanochat project, basically doing ML research autonomously; “autoresearch” as he calls it (there you go, another coinage from the king). The agents ran about 700 experiments, found roughly 20 changes that improved performance, and all of them were real improvements that transferred to larger models. Total gain: 11%. |
AI Supremacy Substack - Ambient Computing via Voice AI is about to Enter its Golden Age in 2027 | I believe once better AI wearable devices are launched next year in 2027 including AI pins (wearables) and smart glasses form factors, it could be a breakthrough year for consumer AI voice experiences. From smart glasses to new kinds of pendants and pins, I expect Apple to dominate. Mid to late 2027 is the time this should really come to the foreground or around 18 months from now |
How We Frame Machines Substack - What Happened When I Told the AI to Stop Helping | The Claude chat was supposed to be brainstorming help — I was stuck on a chapter, a scene involving a character named Ellie whose psychology I’d been circling for three drafts without quite landing. I’d asked for ideas. Claude delivered structural suggestions, tonal variations, and questions worth considering. It was thorough. It was generous. It was making things worse. |
The enterprise launch arrives barely two weeks after Computer debuted for consumers, where it triggered what the company describes as a viral moment: users on social media demonstrated the agent building Bloomberg Terminal-style financial dashboards, replacing six-figure marketing tool stacks in a single weekend, and automating workflows that previously required dedicated teams. Perplexity says more than 100 enterprise customers messaged the company over a single weekend demanding access. | |
ZDNet - Why Moltbook and OpenClaw are the fool's gold in our AI boom | The AI business has become downright crazy. First, OpenAI hired Peter Steinberger, creator of the popular, horribly insecure open-source agent framework OpenClaw. Now, Meta has acquired Moltbook, the viral AI agent social network that also has no security to speak of. This is nuts. |
Computerworld - Microsoft seeks a stay on DoD’s effective ban on Anthropic offerings | Microsoft is urging a federal court in California to temporarily pause the US Department of Defense’s (DoD) effective ban on Anthropic’s AI offerings, arguing that the government’s “supply chain risk” label could have significant knock-on effects for its own defense technology business. |
AI: Reset to Zero Substack - AI: AI Agents increasingly need an Internet of their own. RTZ #1023 | We’re already seeing it with now OpenAI’s OpenClaw, AI Agents running locally with their own agency around the world, on the behalf of their users 24/6. Increasingly needing to work far more efficiently. Beyond using ‘human’ browsers, computers, software tools and applications like file systems, spreadsheets, messaging and other CLI (command line interface) software, all built for humans. Going far beyond open source OpenClaw AI Agents on Mac Minis. |
VentureBeat - The limits of bubble thinking: How AI breaks every historical analogy | AI is different because it performs cognitive work. And if that makes you uneasy, it should. Because if AI can actually think, then a lot of what we’ve built our careers on, like our expertise and our hard-won skills, might not be as defensible as we thought. The junior engineer who spent years developing intuition now works alongside a tool that has it instantly. So does the financial analyst known for their variance analysis. People aren’t completely sure where value actually lives anymore, and that’s terrifying |
The Buzz by Geoff Livingston Substack - What Does Agentic AI Mean? | As I noted, my bias against agentic AI hype is actually a barrier for me. I have to really try and check this contempt at the door, for behind the hype, there may actually be a useful tool that can help a client or me. That’s what matters, not how a company or influencer overhypes and confuses the market. |
AI Supremacy Substack - What is Advanced Machine Intelligence or AMI Labs? | Like many of you I’ve been following the criticism around LLMs by the likes of Yann LeCun, Gary Marcus and many others in the last few years. Yann LeCun, a Turing Award winner and a pioneer of modern AI, has become one of the most prominent critics of the current "LLM-centric" path and his alternative is to me fascinating. Instead of AGI they propose Superhuman Adaptable Intelligence, or SAI. |
It wasn't a finished model or a massive corporate product: it was by his own admission a simple, 630-line script made available on Github under a permissive, enterprise-friendly MIT License. But the ambition was massive: automating the scientific method with AI agents while us humans sleep. | |
Fortunately, there are solutions. Some scientists are building LLMs that can act as guardrails. Yes, adding one LLM to fix the problems of another one seems like doubling the potential for trouble, but there’s an underlying logic to it. These new models are specially trained to recognize when an LLM is potentially going off the rails. If they don’t like how an interaction is going, they have the power to stop it. | |
The takeaway for enterprises as they roll out AI assistants and agents is that they shouldn't assume people will instinctively figure out how to address them. Therefore this is something that needs to be part of the training and change management processes when introducing them — especially while AI agents remain so novel | |
Computerworld - Anthropic sues US government over ‘supply chain risk’ designation | But Anthropic said that its resistance to two items in the government’s contract– autonomous lethal warfare and mass surveillance of Americans–was entirely technical, based on Anthropic testing showing that “Claude cannot safely or reliably perform those functions.” |
But now, just 24 hours after shipping the open source Qwen3.5 small model series—a release that drew public praise from Elon Musk for its "impressive intelligence density"—the project’s technical architect and several other Qwen team members have exited the company under unclear circumstances, raising questions and concerns from around the world about the future direction of the Qwen team and its focus on open source. | |
The Algorithmic Bridge Substack - How Many People Does It Take to Kill a ChatGPT? | After OpenAI announced its Pentagon deal this past weekend, ChatGPT mobile app uninstalls in the US spiked 295%. Downloads dropped 13%. One-star reviews surged 775% in a single day; five-star reviews fell by half. The QuitGPT campaign—a grassroots initiative that started in January 2026—claims 2.5 million signatures. My entire X timeline is people urging others to cancel ChatGPT and switch to Claude. |
The Labs' new technique, Self-Flow, introduces an "information asymmetry" to solve this. Using a technique called Dual-Timestep Scheduling, the system applies different levels of noise to different parts of the input. The student receives a heavily corrupted version of the data, while the teacher—an Exponential Moving Average (EMA) version of the model itself—sees a "cleaner" version of the same data. | |
SiliconANGLE - Microsoft open-sources multimodal reasoning model with 15B parameters | Microsoft compared the algorithm to several similarly sized reasoning models using a set of open-source benchmarks. Phi-4-reasoning-vision-15B scored 17% higher than Google LLC’s gemma-3-12b-it on MathVista_Mini, a benchmark that comprises multimodal math questions. The model also achieved higher scores across more than a half-dozen other evaluations. |
The 15-billion-parameter model, available immediately through Microsoft Foundry, HuggingFace, and GitHub under a permissive license, processes both images and text and can reason through complex math and science problems, interpret charts and documents, navigate graphical user interfaces, and handle everyday visual tasks like captioning photos and reading receipts. | |
Educating AI Substack - When AI Says “This Quote Is Accurate,” You Shouldn’t Believe It | They do not retrieve text the way databases do. They reconstruct language probabilistically, token by token, based on patterns, likelihoods, and semantic approximation. Even when the original document is fully present in the prompt, the model does not perform exact character-by-character comparison. It generates what seems right. And what seems right is often close enough to feel authoritative without being literally accurate. |
ZDNet - Will AI make cybersecurity obsolete or is Silicon Valley confabulating again? | Wall Street observers think there is a real possibility that AI firms' tools will displace the traditional cybersecurity offerings from companies such as Palo Alto Networks, Zscaler, and Check Point Software. A related field, called observability, is also threatened, including firms such as Dynatrace that sell tools to detect system failures. |
A little over a year ago, China’s open source ‘DeepSeek’ up-ended the US AI developer community with its innovations. Now it seems that the now US OpenClaw, with its open source AI Agents running on local (mainly Apple computers), is up-ending the AI/tech developer community in China. Exponentially benefiting and expanding the AI ecosystems in both places. Let’s unpack. | |
It’s a startling mea culpa from Altman - “Good learning experience for me as we face higher-stakes decisions in the future” - and one triggered presumably by a combination of online opprobrium aimed at the company, an uptick in cancellations of ChatGPT, and the rise of Anthropic’s Claude app to the number one slot on the AppStore from being #131 a month earlier. | |
SiliconANGLE - Anthropic makes switching from competitors easier with new transfer memory tool | As Anthropic PBC’s Claude chatbot moves to the top of the app charts after a spat with the Trump administration, the company has introduced a memory import feature that will allow new customers to import their conversations from rival chatbots and start again with Claude. |
To put this into perspective, these models are on the order of the smallest general purpose models lately shipped by any lab around the world, comparable more to MIT offshoot LiquidAI's LFM2 series, which also have several hundred million or billion parameters, than the estimated trillion parameters (model settings) reportedly used for the flagship models from OpenAI, Anthropic, and Google's Gemini series. | |
Given that the Finance role is one of those frequently cited as vulnerable to being automated by AI, Intuit is clearly on the front line of those firms that find themselves caught in the fall-out from the so-called ‘SaaSpocalypse’. | |
The first time my team shipped an agent into a real SaaS workflow, the product demo looked perfect. The production bill did not. A small percentage of sessions hit messy edge cases, and our agent responded the way most agents do: it tried harder. It re-planned, re-queried, re-summarized and retried tool calls. Users saw a slightly slower response, and finance saw a step-change in variable spend. | |
Computerworld - Trump administration bans Anthropic, seemingly embraces OpenAI | Under the plan, according to Axios, the Defense Department would sever a contract, worth up to $200 million, with Anthropic, and require defense contractors and other vendors to certify they are not using Anthropic’s Claude model in work tied to the Pentagon. The administration is allowing a six-month window to give agencies and contractors time to transition to alternatives. |
SiliconANGLE - Satya’s sacrifice: Why agents threaten Office and how Microsoft responds | The real issue is that Nadella’s prediction exposes a vulnerability to Microsoft’s single-user productivity franchise. Specifically, we see agents and a new work surface as the primary interface for knowledge work. Office risks being partially disintermediated – reduced from the place where humans and agents collaborate to a set of file formats that agents can create, edit and orchestrate using open-source engines while users live inside a new work surface – some container that’s more editable than Claude Cowork, and where Word, Excel and PowerPoint are plug-ins |
Diginomica - A little bit of history repeating? Déjà vu lessons for AI code development | The code generation capabilities are intriguing and often capture a lot of software customer interest. The WOW factor on this gets people excited – really excited – but, maybe we should look to the past at previous WOW moments in IT and see if a more considered, premeditated approach to AI is appropriate. |
ZDNet - Is Perplexity's new Computer a safer version of OpenClaw? How it works | On Wednesday, the company introduced Computer, a multiagent orchestration system that harnesses the strongest capabilities from more than a dozen frontier AI models. Currently available only to Perplexity Max users -- and expected to roll out to Enterprise and Pro subscribers in the coming weeks -- "Computer is a general-purpose digital worker," the company wrote in a press release, that "reasons, delegates, searches, builds, remembers, codes, and delivers." |
Computerworld - LinkedIn moves to offer skill validations in the AI era | The Verified AI Skills program unveiled in January involves LinkedIn partnering with AI tool providers to automatically validate and display a user’s proficiency directly in their certification section. The initial partners include Lovable, Replit, Relay.app, and Descript, which will track AI proficiency of candidates using their tools to create AI apps. |
ZDNet - 90% of sales teams use AI agents - but half of them have the same data problem | AI agent adoption requires better data and fewer tools. Sales teams are unifying data and simplifying tech to improve AI and agent outcomes. Sales pros have data concerns, such as manual errors and duplicate data. Others say bloated tech stacks delay their AI initiatives. 84% of teams without an all-in-one platform plan to consolidate tech |
But what Claude did was a real eye-opener. He downloaded the service’s command-line interface and used it to do all the work (except logging in—I had to do that). He couldn’t (yet, I suppose) use the website itself. But a CLI? Child’s play. | |
The company is positioning Red Hat AI Enterprise as a “foundation for AI production” that provides capabilities including AI inference, model tuning, customization, deployment and management tools in a single package. It’s meant to support any kind of AI model in any environment, including the cloud or on-premises. | |
AI Workplace Wellness Substack - The Quiet Rise of AI Fatigue - by Paul Chaney | One of Khare’s most striking observations is how AI changes the nature of work itself. Instead of building things deeply, many engineers now spend their days reviewing AI outputs: editing, validating, correcting. He likened the experience to judging an endless assembly line of code. |
SiliconANGLE - Anthropic slams Chinese AI firms for harvesting data from its Claude chatbot | The process of using data from one AI system to train another is known as “distillation,” and it’s a fairly common technique for developers. But Anthropic’s terms of service prohibit anyone from harvesting Claude’s responses in this way. In addition, they’re meant to prevent its chatbot from being used by anyone in China. |
Marcus on AI Substack - Turns out Generative AI was a scam - by Gary Marcus | None of what Ovide had to say about the overestimation of Generative AI should actually come as a surprise. Generative AI has been inherently unreliable from the start; none of the problems that I warned about over the last half decade has been properly solved. Large language models still hallucinate, and they still make boneheaded errors; they still lack a proper concept of reality. They often produce workslop. A recent survey called The Remote Labor Index found that they could only do 2.5% of human tasks, and that is a massive overestimate, since literally everything that requires physical labor was excluded. |
SiliconANGLE - Why real-time voice AI is harder than it sounds | But even strong models are only part of the equation. Enterprise voice systems must be deployed like infrastructure, and the needs of business buyers are fundamentally different from those of consumers. “It has to have low latency, it has to have high throughput, it has to be reliable, it has to be debunkable, it has to be adaptable and get better over time,” Stephenson said. |
He added, “Enforcement is the hard part. There’s no magic scanner that can reliably detect AI‑generated code and there may never be such a scanner. The only workable model is cultural: require contributors to explain their code, justify their choices, and demonstrate they understand what they’re submitting. You can’t always detect AI, but you can absolutely detect when someone doesn’t know what they shipped.” | |
Despite the promise of smaller models, many of the tools on display in the exposition hall at Developer Week were designed to facilitate access to leading AI models such as those from OpenAI Group PBC, Anthropic PBC and China’s DeepSeek. The developer community is looking closely at the merits of both small and large models, a situation that will likely become clearer over the coming year. | |
ZDNet - Is an AI subscription worth it? How to choose your premium chatbot plan - and what not to do | All have surprisingly capable free plans. ChatGPT, Gemini, and Grok have $8-per-month basic plans that provide a bit more interaction capability but use lesser-capability AI engines. All but Grok offer $20-per-month standard plans. These are the sweet spot for price and performance. Grok's equivalent, called SuperGrok, is $30 per month instead of $20. |
Normally, when big-name talent leaves Silicon Valley giants, the PR language is vanilla: they’re headed for a “new chapter” or “grateful for the journey” — or maybe there’s some vague hints about a stealth startup. In the world of AI, though, recent exits read more like whistleblower warnings |