AI Agents Are Getting Ready for the Office

Noted

5–7 things worth understanding, Today by Stackari

· October 8, 2026

Read online | Sign up | Advertise

Your Daily Note,

Agents are moving into business workflows, computers are being built around local AI, and the web is starting to figure out how to deal with software that acts like a user. Meanwhile, AI is finding its way into healthcare, gaming, advertising and even hotel discovery.

In today’s Noted:

AI agents are getting ready for the office
Your next PC might be built around AI
Muse is becoming a more useful AI agent
ChatGPT can now build the interface, not just the answer
Websites are starting to prepare for AI agents

NOUS RESEARCH

this one’s big →

AI agents are getting ready for the office

We noted:

Nous Research has raised $90 million in Series B funding at a $1.5 billion valuation and is taking its Hermes AI agent into the business market. The company says Hermes has already been downloaded more than 22 million times since its February launch. Nous is turning Hermes from an open-source AI agent into a product that companies can actually deploy across their teams.

Companies can give employees access to AI agents that can work through multi-step tasks rather than simply answer questions.
Connected tools: Hermes can work with files, browse the web, run code and interact with programs on a computer.
Team controls: the business versions include controls around token usage and allow teams to share agent customizations.
Multiple models: Hermes can route tasks between different AI models based on factors such as cost, speed and output quality.
Persistent context: users can connect external databases so the agent can retain preferences and project information across sessions.

Why it matters

Hermes is moving toward the kind of AI where you give it a goal rather than a single prompt.

Instead of: “Write me a sales email.”

The longer-term experience is closer to: “Find these prospects, research them, organize the information, draft the outreach and prepare everything for me.”

The agent still needs oversight, especially when it is interacting with real systems, but the direction is clear: more AI work happens between the prompt and the final result.

MICROSOFT + NVIDIA

Your next PC might be built around AI

We noted:

Microsoft and Nvidia have introduced new Windows hardware built specifically for demanding local AI workloads. The headline device is the Surface Laptop Ultra, built around Nvidia’s RTX Spark platform with up to 128GB of unified memory. Microsoft says it can run AI models exceeding 120 billion parameters locally. This isn’t just another faster laptop. Microsoft is trying to make the PC itself a serious AI computing platform.

Large models can run directly on the machine instead of sending every request to the cloud.
Developers can test models, build agents and run coding workflows locally.
Hybrid computing: Windows can move between local and cloud computing depending on the workload.
Agent security: Microsoft is introducing Execution Containers to isolate AI agents and control what they can access.
AI hardware: the new Surface devices combine Nvidia’s GPU and CPU technology with large amounts of unified memory.

Why it matters

The future PCs may increasingly handle AI tasks locally. Things like coding agents, image generation, private document analysis and other AI workloads could happen on the device without everything being sent to a remote server.

That could make AI faster and potentially more private, while reducing how much cloud computing is required.

META MUSE

sneaky useful →

Muse is becoming a more useful AI agent

We noted:

Meta has brought its Muse AI assistant to iPad only a month after its mobile launch. Muse can now connect to: Canva, Dropbox, Figma, QuickBooks, GitHub, Zoom, Asana, Klaviyo, Granola and Notion are among the supported tools. Muse is designed to take a goal and work through it rather than simply answer a question.

More devices: Muse now has a dedicated iPad experience alongside its mobile apps.
More work apps: integrations let Muse work with tools businesses already use.
Business workflows: Meta says Muse can help with things such as marketing plans, client proposals and product pages.
Shopping: Muse is also being positioned as an agent that can help people discover and purchase products.
Agent infrastructure: Muse runs inside Meta’s own secure virtual environment so it can act on a user’s behalf.

Why it matters

The experience Meta is building is less “ask an AI a question” and more “give an AI something to take care of.”

For example, instead of asking for ideas for a marketing campaign and then opening Canva yourself, an agent could eventually take the brief, work across your connected tools and prepare the pieces for you.

The more integrations Meta adds, the more useful Muse becomes because it has more places where it can actually do something.

OPENAI

updates →

ChatGPT can now build the interface, not just the answer

We noted:

OpenAI has introduced Intelligent UI alongside GPT-6. Instead of returning every answer as a block of text, ChatGPT can now decide whether a question is better answered with text, visuals or an interactive tool. That means a conversation can produce something you can actually use inside ChatGPT, rather than something you need to take elsewhere.

Interactive answers: responses can include charts, graphics, buttons, forms and interactive elements.
Tools on demand: ask for something like a calculator, bill splitter or game and ChatGPT can create it directly in the conversation.
Visual explanations: complicated ideas can be turned into interactive diagrams where you can change inputs and see what happens.
Adaptive layouts: GPT-6 chooses the format based on the task. A comparison might become a side-by-side interface while a simple question can remain plain text.
Progressive answers: GPT-6 can start responding while it continues thinking or using tools, then add more findings as they become available.

Why it matters

The bigger change is that users don’t necessarily need to know what tool to use first. They can describe what they want and let ChatGPT decide how to build the experience. This may increase the reliance on ChatGPT than switching models.

THE AGENTIC WEB

New changes →

Websites are starting to prepare for AI agents

We noted:

AI agents are increasingly expected to browse websites, search for information and eventually complete actions on a user’s behalf. But websites were largely designed for humans and search crawlers, not software that can independently click, read and act. A new group of web standards is emerging around this problem. One example is the Agent Web Protocol, which proposes a machine-readable agent.json file that tells AI agents what a website can do, how it can authenticate and which actions are available.

Declaring capabilities: websites can explicitly tell agents what actions they support.
Structured actions: instead of an agent figuring out a website by scraping the interface, a site can describe actions such as search, booking or purchasing.
Authentication: the protocol includes ways for sites to declare supported authentication methods and permissions.
Agent discovery: an agent can check a standard location for the site’s capabilities instead of guessing how the website works.
Better control: websites can decide what agents are actually allowed to do rather than simply treating all automated traffic as the same.

Why it matters

The web has spent decades optimizing for humans and search engines. Terms like AEO, SEO, GEO reflect that history.

AI agents create a third major type of visitor.

If agents become the way people increasingly interact with websites, websites may need to become machine-readable and machine-actionable, not just human-readable.

Product Worth Knowing

I

IrisGo

Show the AI how you do a task once, then let it run the workflow for you. It’s aimed at repetitive work that normally requires manually moving between steps and tools.

V

Velozity

A shared workspace where people and AI agents can work together rather than keeping AI in a separate chatbot window.

G

GenPage 3.0

Help businesses create landing pages and then personalize and optimize them for different campaigns or audiences, rather than sending everyone to the same generic page.

Everything else in AI

The goal: Anyone can make a game with AI

Google has launched Playground, an experimental AI platform where people can create, play and share custom browser games without writing code. Users describe what they want through text prompts and Playground generates the experience. Google says more professional-grade tools are coming through its Unity Spark work.

AI flags patients who need a closer look

Healthleap raised $38M to scale its AI system that identifies high-risk hospital patients who need urgent intervention or closer monitoring. The software analyzes patient data to flag deteriorating conditions before they become critical, essentially automating the triage process that hospitals currently do manually. The company is testing the system for conditions including malnutrition, delirium, aspiration pneumonia and pressure ulcers. The company says its system is already deployed across more than 50 hospitals.

NVIDIA’s AI models are getting better at Olympiad-level problems

NVIDIA says its Nemotron models achieved gold-level results at both the International Olympiad in Informatics and the International Mathematical Olympiad after being specifically trained and optimized for these types of problems. NVIDIA is fine-tuning models for difficult mathematical and programming problems. The systems combine model reasoning with specialized training techniques.

ChatGPT is becoming a college planning tool

OpenAI is adding tools designed to help students with college planning, including organizing applications, deadlines and financial-aid information. Students can organize the different parts of the application process. ChatGPT can help break a large planning task into smaller steps. The tools are designed to help students research and manage information rather than simply generate an essay.

Radisson hotel discovery

Radisson is bringing hotel discovery into ChatGPT, giving travelers another way to search for and explore accommodation through an AI conversation. Travelers can describe what kind of hotel or trip they want. AI can help narrow down relevant options. The experience brings hotel discovery closer to the point where someone is already making a travel decision.

Melius is moving deeper into the ad-making process

Melius, founded by former Ramp engineers, has raised $20 million for an AI platform focused on advertising creative and campaign production. The company is targeting the repetitive parts of advertising where teams constantly need new ideas, images, videos and campaign variations. Side note: Ads generation is not new; we are excited to see who can generate the best idea and turn them into great content.

Worth your weekend

→ Read our blog: Jasper Alternatives for Small Teams 2026 → View New Tool on Stackari → Missed a note? Browse the archive

See you tomorrow
same time, different note.

~ Stackari

P.S. If you found this useful, pass it along to someone curious.

You signed up for Stackari’s newsletter. Glad to have you here.

Stackari · More news · Privacy · Unsubscribe

October 08

AI agents are getting ready for the office

Why it matters

Your next PC might be built around AI

Why it matters

Muse is becoming a more useful AI agent

Why it matters

ChatGPT can now build the interface, not just the answer

Why it matters

Websites are starting to prepare for AI agents

Why it matters