Anthropic launched a beta version of its Claude Tag feature for Enterprise and Team tiers, shifting its chat model into shared Slack channels. Moving away from traditional isolated chat boxes, users pull the artificial intelligence model into active group threads by typing @Claude. The integration allows any team member in the channel to delegate a task, review the model’s outputs, and pick up the discussion thread from a previous point. This structural shift follows a US$65 billion Series H funding round that brought Anthropic’s post-money valuation to US$965 billion, positioned above rival OpenAI’s US$852 billion mark. Following a confidential S-1 filing for an initial public offering, market competition for business software placement remains tight. Data from corporate expense platform Ramp’s May 2026 AI Index indicates Anthropic’s enterprise adoption rate reached 34.4%, passing OpenAI’s 32.3% footprint. Standard generative software requires enterprise employees to move data between team chats and separate browser instances. Anthropic aims to reduce this back-and-forth movement by restructuring workplace AI agents to work in multiplayer environments. “Instead of a private back-and-forth, Claude Tag shows up in the open,” stated Rob Seaman, general manager of Slack, regarding the operational mechanics of the application. This shared visibility alters how context is tracked inside an organisation. Because Claude Tag logs its task status directly inside the communication window, multiple employees can monitor the live execution steps. The system tracks ongoing information from its active channels to build a contextual background. This automated history tracking limits the need for team members to continuously retype foundational company data or project scopes. The technical foundation for this channel integration relies on Anthropic’s Opus 4.8 engine. When assigned a request, the model divides the operation into sequential execution phases and utilises connected corporate databases, tools, and code repositories to complete the work. The primary operational difference for these workplace AI agents is their capability to function asynchronously without real-time human prompting. If a network administrator activates the tool’s “ambient” configuration, Claude Tag monitors threads and tracks tasks autonomously. The agent checks inactive text threads, signals priority notifications from integrated software extensions, and tracks unresolved assignments across multi-day intervals. Cat Wu, head of product for Claude Code, noted that the change centres on user configuration rather than completely new logic. “The form factor of being able to tag it the same way that you would a coworker is really powerful,” Wu told Reuters. Wu explained that connecting her personal Claude Tag agent to her email archive allows the system to analyse incoming communications, categorise urgent entries, and send immediate alerts inside Slack. Internal reporting from Anthropic shows that automated code generation has altered engineering activities, with the firm’sinternal product group creating 65% of its code through its private version of Claude Tag. Beyond software development, the vendor targets non-technical office workforces. Early customer implementations focus on querying database metrics, parsing analytics data, and processing internal IT support tickets. This expansion of background agent operations requires a distinct security infrastructure to protect proprietary information. To restrict data access to approved departments, system administrators must establish scoped Claude identities. All localised memories and tool integrations are confined strictly to specific channels authorised by the IT department. Additionally, management portals offer full tracking logs of user queries alongside specific organisational caps to regulate monthly token costs. Frankly, moving generative tools from individual sandboxes into persistent corporate communication channels presents distinct operational trade-offs. The clear upside is the optimisation of routine knowledge work. By centralising information logs directly inside active threads, companies can lower task friction, capture context across changing project teams, and reduce the time spent on manual codebase tracking or database updates. However, delegating cross-app workflows to background agents introduces significant structural risks for IT departments. Permitting automated systems to read chat histories, connect to email accounts, and modify central code repositories expands an organisation’s internal data-exposure risks. If access boundaries are misconfigured, sensitive proprietary context could cross into unapproved channels.
Samsung Electronics is expanding employee access to ChatGPT Enterprise and Codex, giving staff wider use of AI tools for technical and non-technical work. According to OpenAI, the deployment covers all Samsung Electronics employees in Korea and all Device eXperience employees worldwide. The DX division includes smartphones, consumer electronics, and home appliances. Samsung plans to use the tools in software development, marketing, product development, manufacturing, and other business functions. The tools will support tasks such as information search, document drafting, idea development, data interpretation, and code-related work. The rollout comes three years after Samsung restricted employee use of generative AI tools over data-security concerns. In 2023, the company limited the use of ChatGPT and similar tools after concerns that sensitive internal information had been uploaded to an external AI platform. The new deployment gives employees access to ChatGPT Enterprise, which includes controls for data protection, user access, and security management. OpenAI said the enterprise version allows organisations to manage users, apply access controls, and use AI tools within internal security requirements. Samsung’s earlier restrictions applied to employee use of ChatGPT and similar generative AI tools. The new rollout gives employees access through an enterprise product with data protection and access controls. Samsung has not limited the deployment to a single business unit or technical group. OpenAI said the tools will be used across a broad range of functions, including technical and non-technical teams. OpenAI said ChatGPT can support knowledge-based tasks such as searching for information, analysing material, drafting documents, developing ideas, and interpreting data. Codex will be used for software-related tasks such as writing, reviewing, and debugging code. OpenAI said the tool is also being used for internal tools, websites, software prototypes, and automated workflows. OpenAI said Codex can also support non-technical teams in day-to-day work, including by helping employees create internal tools and automated workflows. OpenAI said Codex now has more than five million weekly users across technical and non-technical workflows. In Korea, weekly active users of Codex have grown nearly 800% since February 1, 2026, according to the company. Harrison Kim, general manager of OpenAI Korea, said the agreement is one of OpenAI’s largest enterprise deployments. He said Samsung is using AI across teams and functions rather than limiting it to specific departments. In October 2025, Samsung said it would work with OpenAI as a strategic memory partner for the Stargate AI infrastructure initiative, with OpenAI’s memory demand projected to reach up to 900,000 DRAM wafers per month. Samsung SDS also entered a potential partnership with OpenAI to jointly develop AI data centres and provide enterprise AI services. Samsung said the agreement would allow Samsung SDS to provide consulting, deployment, and management services for businesses integrating OpenAI models into internal systems. Samsung SDS also signed a reseller partnership to offer OpenAI services in Korea. Under that arrangement, Samsung SDS said it would support Korean companies adopting ChatGPT Enterprise and other OpenAI services. Reuters reported that Samsung Electronics and SK Hynix had signed letters of intent to supply memory chips for OpenAI’s Stargate project. The report said the two South Korean chipmakers together account for about 70% of the global DRAM market and nearly 80% of the high-bandwidth memory market. High-bandwidth memory supports fast data movement between memory and processors in AI systems. Reuters reported that OpenAI’s chip demand for Stargate may reach 900,000 wafers per month, citing South Korea’s presidential office. Samsung said its semiconductor businesses would support OpenAI’s demand with advanced memory solutions. The company also said its affiliates were exploring broader work with OpenAI in areas including data centres, enterprise services, and AI infrastructure. Deloitte’s 2026 State of AI in the Enterprise report found that 66% of organisations reported productivity or efficiency gains from enterprise AI adoption. The same report found that 53% reported improved insights and decision-making. A Bpifrance survey reported by Reuters found that 77% of 534 French mid-sized company heads said their firms used generative AI, but only 17% of those using it reported time savings. Samsung has identified use cases across document work, information analysis, coding, product development, marketing, and manufacturing. The deployment gives employees access to ChatGPT Enterprise and Codex for those tasks under a company-wide agreement. OpenAI has also announced other partnerships in Korea.
- RFIC design is a complex “dark art” that limits progress in wireless technologies like 5G, autonomous vehicles, and satellite communications. - Princeton researchers use reinforcement learning and inverse design to rapidly create RFICs from scratch. - Diffusion models rapidly generate novel or human-interpretable RF layouts, achieving record performance and drastically reducing design time. - Future progress needs large, shared chip design datasets and open ecosystems so AI can learn universal electromagnetic and circuit behaviors. Take a moment and try to imagine your life without the wireless advances of the past three decades. Have you lost your luggage? What a shame AirTags have not been invented. The airline representative has promised to call with updates, so settle in for a long wait by the kitchen telephone, because there are no affordable cellphones. You’ll be stuck listening to whatever is on the radio while you wait, because there are no streaming services. That’s not even to speak of all the movie plots that would have been ruined. This is just a tiny sliver of how wireless technology makes itself felt in your day-to-day existence. The effects it has had on supply chains, infrastructure, and how the economy runs have been world-altering. None of it would be possible without the radio-frequency integrated circuits that allow all our devices to unobtrusively send and receive information. Now imagine what the further evolution of this technology will bring: Wide-spread autonomous vehicles, quantum communications, 6G mobile service and satellite communications. Continued momentum will depend on newer and more advanced versions of today’s RF chips. But there’s the rub. Whereas the design of most of the world’s computing chips has been standardized into its own science, RF design has remained stubbornly in the realm of art. A dark art, even, that is mastered only through years of experience. As any sorcerer will tell you, the dark arts keep their own schedule. And that schedule is impeding progress not just in RF chip design but in every other technology that depends on it. About seven years ago, in the wake of AlphaGo’s victory over world Go champion Lee Sedol, my students at Princeton and I began to wonder: Could AI be taught this art as well? Recent successes suggest that, to a large extent, it can. Over the last few years, our group and other leaders in the field have started to develop machine-learning-driven algorithmic methods for designing RFICs. Some of the resulting chips look more like modern art than circuit layouts. Yet in many cases, the physical prototypes bested state-of-the art circuits in terms of performance. The real achievement, however, is that it took the AI orders of magnitude less time to conceive a working design than it would a human designer. This is not about one or two RF chips. AI-enabled design could be the future of all RF design, and maybe much more. So why do these chips all have to be crafted by hand? Why aren’t RFICs designed with an algorithmic synthesis process, much as CPUs and GPUs are? The design of RFICs is an exercise in engineering across multiple physical domains. Maxwell’s equations, operating across different spatial and temporal scales, govern how electromagnetic fields interact with active and passive devices that must be carefully codesigned for the chip to function. Alongside these are the laws of thermodynamics, which determine how heat is generated and removed during operation, as well as the mechanics of thermal expansion and contraction that dictate how reliably the chip and its packaging survive temperature changes. Simultaneously accounting for all the physical constraints these impose makes the design space almost impossibly large. Every decision involves complex priorities that often compete with one another, preventing the optimization of any of them. To better understand the issue, let’s walk through the steps involved, after which you’ll better understand why a single new chip design takes years and tens to hundreds of millions of dollars. Most of the area of radio-frequency integrated circuits is dominated by complex electromagnetic structures. Human-designed RFICs, like this broadband power amplifier [1], start with templates and follow a symmetric, understandable pattern. But freed from the constraints of human-designed templates and the need for humans to even understand the rationale of electromagnetic structures, power amplifier ICs [2–5] and low-noise amplifiers [6] can take on truly wild-looking yet efficient designs. SENGUPTA LAB Let’s say you’re an engineer assigned to design a new 28-gigahertz power amplifier for a 5G-millimeter-wave handset. (This is the type of RFIC that boosts the 5G signals on your phone and transmits them to the antenna where they can be picked up by a distant base station). Where do you start? RFIC design has some features in common with house building.
AI is booming. New use cases are emerging each day. To capitalize on the technology’s potential, enterprises require data at scale. In many cases, though, the relevant information is blocked or unstructured, which limits its use by AI models. To understand this challenge, consider the foundation of the web itself. The web was not designed for the automated discovery and retrieval that new AI applications demand. Overcoming this inherent design constraint requires infrastructure. The next frontier in AI may depend on a new web data infrastructure layer that can enable models to discover and map this ever-expanding digital realm. This layer must be able to navigate hundreds of millions of existing web domains and billions of new URLs created each week, delivering real-time information and overcoming technical barriers. “The data suggests there’s far more data out there,” says Or Lenchner, CEO of Bright Data, a web data collection platform. “Think of the universe: It’s out there, but you don’t know what you don’t know.” While early AI breakthroughs were driven by scaling training data and model size, organizations are now encountering a fundamental bottleneck: They need to keep pace with the dynamic, unstructured, and constantly evolving nature of web data in order to ground outputs in current and verifiable information. AI performance increasingly depends not just on model architecture but on a system’s compute, networking, retrieval, and data engineering capabilities—that is, the system’s ability to quickly and reliably retrieve data that is fresh, relevant, and trustworthy. Traditional model training relies on snapshots of information collected at a particular point in time. Training AI on such static data is no longer sufficient. To track fluctuations such as competitor pricing, consumer sentiment, and market trends, companies need a constant feed of new information, pulling data in real time along with relevant context. Their infrastructure must therefore be able to handle millions of simultaneous interactions across websites that vary by geography, language, format, and access rules. “If it can’t retrieve real-time information, it lacks context,” Lenchner says. “In a business setting, that’s not acceptable anymore. Stale answers lead to bad decisions and disappointed consumers.” Speed is not merely a matter of convenience; it’s a matter of necessity. Today’s organizations operate in environments where prices, inventory, markets, security threats, and customer behavior change continuously. Delayed data retrieval can reduce the usefulness of an otherwise sophisticated model. Using live, high-quality web data can also reduce AI hallucinations because the model has a more relevant knowledge base. This builds user trust. In fact, one survey found that 56% of AI practitioners said businesses need access to real-time web data to improve trust in AI outputs. To ensure the model runs efficiently and effectively, the information must also be pared down to the appropriate essentials. Despite the introduction of retrieval-augmented generation (RAG), where models pull in external data at the moment of a query, many AI systems still struggle to deliver outputs that are current, contextually relevant, and trustworthy in operational settings. According to Gartner, 60% of AI projects that are not supported by AI-ready data—accurate, structured, organized, and contextualized—will be abandoned by the end of the year. This is because large-scale retrieval alone does not solve the problem. As Lenchner puts it, “You need to retrieve data at scale, but also in real time. Latency becomes an issue because of the end user who is waiting for the output.” Accessing fresh, AI-ready data at scale introduces technical and structural challenges. In practice, many enterprise systems combine public web retrieval with APIs, licensed datasets, and proprietary internal data in their AI applications. Integrating these fragmented sources into a timely and usable knowledge layer requires specialized capabilities. Some research has found that 97% of AI organizations depend on real-time web data infrastructure, but 90% feel boxed in by various restrictions. Companies are increasingly developing technical approaches to navigate these constraints. Lenchner draws this metaphor: “Think of the trained model as intelligence and relevant data as knowledge. A powerful intelligence layer sitting on top of a hollow knowledge layer is like a genius who knows nothing—useless in practice. Intelligence and knowledge have to come together.” A new layer of web data infrastructure can address this developing need for stronger AI inputs by enabling discovery of data, real-time access, and tailoring to a specific context.
Mistral just dropped OCR 4, and the thing’s built on a tighter feedback loop than their previous version. Instead of just scaling up data, they rewired the token‑alignment stage so the model can keep track of layout cues while it parses characters, which ends up trimming the error cascade you usually see with PDFs and slide decks.
In a blind test they ran against a handful of commercial OCR services, OCR 4 came out on top in roughly 72 percent of the cases. The win wasn’t about raw speed; it was the consistency of getting tables, footnotes and mixed‑language snippets right without the usual jitter.
What’s neat is they’re packaging it as a plug‑in for existing document pipelines, so you can swap it in without re‑architecting your workflow. If you’ve been wrestling with messy scans, this might finally give you a clean pass‑through.
So Anthropic’s just dropped a little Slack‑sidekick they call Claude Tag. Instead of opening a separate UI, you just @Claude in any channel, drop a prompt, and it starts chipping away at the task—whether that’s drafting a spec, answering a question, or actually writing code. The magic is that the bot lives right inside the flow you already use, so there’s no context‑switching.
What’s wild is how much it’s already handling for the product team. They say roughly two‑thirds of the code they ship now comes from Claude Tag’s suggestions, with engineers reviewing and polishing the output. It’s not that the AI is taking over; it’s more like a pair programmer that never sleeps, surfacing snippets the moment you need them.
Internally, the tool is wired to pull in the same model that powers Claude, so it has the same reasoning depth. The integration hooks into Slack’s message API, parses the request, runs the model, and drops the result back into the thread. Because it’s just another message, you can iterate instantly—edit the prompt, ask for tweaks, or ask for a different language, all without leaving the chat.
The broader idea is to let teams keep their collaboration space as the single source of truth, while the AI does the heavy lifting behind the scenes. It feels like having a quiet, super‑focused teammate who’s always ready to draft a function or explain a concept, right where the conversation is happening.
Language models may write cleaner prose than most humans, but ask one for 100 arguments on a topic and they'll all cluster together. Human reasoning is far more diverse, says Pangram CEO Max Spero, and that's what might give AI away. The article Pangram CEO says language models give themselves away by making the same arguments appeared first on The Decoder.
Arnaud Fournier is basically saying the whole “DeployCo” thing is less about selling a product and more about slipping AI engineers straight into a company’s existing teams. He sees the engineers as translators, taking the raw output of models like Codex and shaping it into the specific workflows that each business already runs. That hands‑on approach, he argues, creates a feedback loop where the quirks customers hit become data points that feed right back into model training.
What’s surprising is how fast Codex has scaled. He points out that the number of active developers using the tool has multiplied several times over the past year, and that growth isn’t just raw usage—it’s the depth of integration. Teams are now building entire internal tools on top of Codex, not just sprinkling snippets of code here and there. That depth, he says, is what fuels the next round of improvements.
On the pricing side, Fournier notes that the cost per unit of AI work has slipped dramatically because the models are getting better at reusing context and because the infrastructure is being optimized for scale. In practice, that means a company can run more queries for the same budget, or achieve the same output with fewer compute cycles. The drop isn’t just a market trend; it’s a direct result of tighter loops between deployment and model refinement.
Finally, he brings up the ROI question that keeps executives awake. Instead of promising vague “transformative” outcomes, he pushes for concrete metrics: how many hours of developer time are saved, how many bugs are avoided, how quickly a new feature can ship. The idea is to let the numbers speak for themselves, showing that the value of AI isn’t a buzzword but a measurable lift in productivity.
OpenAI is adding custom hardware to its tech stack. The "Jalapeño" chip, developed with Broadcom, is tailored for large language model inference and is set to run at scale by late 2026. The article OpenAI and Broadcom unveil "Jalapeño," a custom chip built for LLM inference appeared first on The Decoder.
So I was reading about this new approach to anchor detection for retrieval augmented generation, and what caught my attention is how they're using parallel detectors to filter out irrelevant information. Essentially, they're running multiple detectors at the same time to identify the most relevant parts of a document, and then they're using a large language model to make the final call.
This approach is interesting because it's trying to balance the trade-off between precision and recall. By using multiple detectors in parallel, they can cast a wider net and catch more relevant information, and then the language model can help refine the results and make sure they're accurate.
It seems like this method is particularly well-suited for working with structured tables and other types of formatted data. The detectors can look for specific keywords and phrases, and then the language model can help understand the context and make sure the results are relevant.
What's also notable is that this approach is using a combination of traditional retrieval methods, like keyword search and table of contents analysis, along with more modern techniques like embeddings and large language models. It's not relying on any one method, but instead using a combination of approaches to get the best results.
Send this story to anyone — or drop the embed into a blog post, Substack, Notion page. Every play sends rev-share back to storyflo · A.I..
We’ve simplified responses to 👍 / 👎. Past comments are archived but no longer visible.