Meta Muse Wonder, AI Agents & New AI Startups

Noted

Your daily brief in one short note from us to you.

· October 13, 2026

Your Daily Note,

Meta is making another push for the top tier of AI models. Elsewhere, startups are applying AI to architectural blueprints, personal assistants are learning to listen for context, and businesses are looking for ways to automate communications and marketing.

But as AI becomes more capable, the questions are changing too. Who is responsible when it acts? What happens to the information it hears? And who should set the rules?

In today’s Noted:

–Meta just released its first frontier AI model

–This AI startup wants to turn a sketch into building blueprints in minutes

–An AI assistant wants to listen in the background, not wait for a prompt

–This startup wants to bring senior-level PR expertise to every company

–Microsoft says the UK is well placed to lead global AI regulation

METAthis one’s big →

Meta just released its first frontier AI model

Image source: research.meta.ai

We noted:

Meta has released Muse Wonder and Muse Wonder Full, two larger and more capable AI models from its Superintelligence Lab. The release marks a new step in Meta’s effort to compete with leading frontier-model developers after previously focusing on Muse Spark, a smaller model designed for faster responses.

The new models are intended to handle more complex tasks, including coding, and Meta is putting particular emphasis on their ability to remember and use information about a user over time.

•Launching two versions. Muse Wonder is available through Meta’s Model API, while Muse Wonder Full is reserved for Muse Code subscribers rather than being served through the API.

•Competing on complex tasks. Meta’s model card reports benchmark results comparable to Fable 5.1 and GPT-6 Astra. These are company-reported comparisons, and benchmark results do not necessarily predict performance across every real-world task.

•Building longer-term memory. Meta highlights a capability it calls “active memory,” designed to retain and synthesize useful information across days, weeks and months.

•Making the models available to developers. Meta lists API pricing for Muse Wonder at $4 per million input tokens and $20 per million output tokens, putting it in a similar price range to some other high-end models.

•Connecting the models to its agent. The Wonder models will also power Meta’s Muse agent, tying model improvements to its broader personal-assistant ambitions.

Why it matters

For developers, a stronger model could improve applications that need to work through complicated instructions, write code or retain context across longer tasks.

Muse Wonder is an attempt to close that gap. Its focus on memory also points toward a broader shift in AI: assistants are being designed not just to answer a single prompt, but to build useful context across many interactions.

BLUEPRINTS AIworth a look →

This AI startup wants to turn a sketch into building blueprints in minutes

We noted:

Blueprints AI is applying generative AI to a field where much of the work is technical, repetitive and tightly constrained by building codes.

The startup was founded by architect and contractor Al Dram and data scientist Coco Cao. Its software can take a two-dimensional sketch or PDF and generate architectural construction documents. The company says its newer version can create coordinated plans across multiple building disciplines.

•Generating more than a floor plan. The second-generation product can produce plans for architecture, structural design, mechanical systems, electrical work, plumbing and interior design.

•Keeping plans coordinated. The company says a change to one part of a project can update the wider set of documents, with checks against applicable local building codes.

•Training on approved drawing sets. The founders say their models were trained using eight years of approved drawings from Dram’s existing design and construction business.

•Expanding beyond a single building discipline. The startup is working on a third version that will add structural analysis and electrical panel design.

•Keeping licensed professionals involved. The company says architects must still review the generated documents and apply their professional seal before permit submission.

Why it matters

The first-generation software costs between $3,200 and $8,500 per month, depending on the plan. The newer version is designed to take on more of the repetitive drafting process, although the company’s claims about speed and code compliance still need to be assessed against individual projects and jurisdictions.

Architecture and engineering firms could use tools like this to produce an initial set of drawings faster, update related documents when a design changes and spend less time on repetitive drafting. Developers and property owners may also find it easier to explore early design options before committing to a full project.

However, building codes vary by location, and generated documents still need professional review. “Permit-ready” should not be mistaken for guaranteed approval by every local authority.

AMBIprivacy check →

An AI assistant wants to listen in the background, not wait for a prompt

We noted:

Most AI assistants wait for someone to open an app and ask a question. Ambi wants to make assistance more proactive by capturing context from conversations and helping users while events are happening.

The startup was founded by Peter Mo, a former Plaud executive, and announced $26 million in funding across multiple rounds. It is releasing apps for iPhone, Mac and Apple Watch, with plans for a dedicated hardware device.

The privacy question: Ambi says ambient capture is off by default, can be switched off at any time, and that it does not store recordings. It also says data is encrypted, although TechCrunch noted that the company did not specify the encryption method.

There is a separate issue when other people are present. Someone using an ambient-listening assistant may capture conversations involving colleagues, clients or friends who have not been told that the technology is active. Ambi acknowledges that users need to disclose this, while saying the wider industry has not fully solved the problem.

•Capturing context throughout the day. Users can choose to let Ambi listen beyond scheduled meetings, rather than activating it only for a specific call.

•Surfacing information during conversations. The agent can draw on a user’s calendar, meetings, knowledge base and web information to offer relevant suggestions while a discussion is taking place.

•Starting work before a meeting ends. Rather than waiting until a transcript is processed, Ambi wants its agent to help users act on information in real time.

•Moving between devices. A task started on one device can continue on another.

•Planning a more capable device. The company says it intends to build hardware with a screen, speaker, camera and microphones to capture context and display information.

Why it matters

The promise is an assistant that can remind someone of a relevant detail during a meeting, help find information without interrupting the conversation or begin follow-up work while context is still fresh. But the usefulness of that experience depends on more than recognition accuracy. People need to understand when capture is active, what is retained and who can access the resulting information.

AI companies are competing to make assistants more personal, and the more context an assistant has, the more relevant its suggestions may become. But gathering that context can create privacy risks that do not stop with the person who installed the app. The challenge is to make proactive assistance useful without making continuous capture the default expectation of everyday life.

HONEYJARfor founders →

This startup wants to bring senior-level PR expertise to every company

We noted:

Public relations is not simply a matter of writing a press release and sending it to journalists. Good communications teams decide what is worth announcing, when to speak, which reporters are relevant and whether a company is ready for attention in the first place.

Honeyjar is building AI tools around that broader communications workflow. The startup was founded by Michelle Masek and Nadia Jamshidi, who have decades of experience in public relations, in-house communications and agency work.

•Finding relevant journalists. The platform can identify reporters and help build media lists for a particular announcement or company.

•Monitoring news and social channels. It can scan coverage, track mentions and surface opportunities for a company to contribute to a story.

•Preparing communications materials. The tools can help draft announcements, pitch emails and social posts in a company’s voice.

•Finding speaking opportunities. Honeyjar also aims to identify events and opportunities where founders or company representatives could share their expertise.

•Building strategy into the workflow. The founders say they are working to incorporate the judgement of experienced communications professionals, including through a broader initiative called Comms Lab.

Why it matters

Honeyjar says its waitlist attracted 10,000 signups in its first 24 hours. TechCrunch reports that beta testers include Nextdoor, Coco Robotics, Bloomberg Beta and communications agencies Mission North and Full Picture. The company says users remain responsible for sending pitches and publishing content, and its founders say the tool should sometimes advise a company that it is not yet ready for media coverage.

Small businesses and early-stage founders may be able to identify relevant journalists, organize outreach and prepare more thoughtful announcements without a full-time communications team. Experienced PR professionals could use the platform to handle research and administration, freeing up time for strategy and client relationships.

MICROSOFTpolicy watch →

Microsoft says the UK is well placed to lead global AI regulation

We noted:

Microsoft president Brad Smith is calling for stronger rules governing AI. His argument is that AI is becoming important enough to national security and everyday life that voluntary promises from technology companies are not sufficient on their own.

Smith argues that the UK is particularly well positioned to help lead international AI safety efforts, partly because its AI Security Institute has been able to test major models before their public release.

•Independent safety evaluations. AI developers should face outside scrutiny of model capabilities and risks before releasing their products.

•Clearer incident reporting. Smith has argued for legal requirements to report security incidents involving AI systems, including cases where agents act in unintended ways.

•Controls for serious failures. He has spoken positively about mechanisms that can pause or slow an AI system when people lose control of its behaviour.

•Rules that go beyond voluntary commitments. Smith argues that companies acting responsibly should not be the only ones following safety expectations. Rules need to apply to less responsible operators too.

•International coordination. His position is that the UK can help bring governments and AI companies together around shared standards.

Why it matters

The immediate impact is unlikely to be a single new rule for every AI tool. The broader debate could shape future requirements around testing, transparency, incident reporting and the deployment of high-risk systems. For businesses adopting AI, clearer standards could eventually make it easier to understand what evidence to request from vendors before using their products.

Smith’s comments are a policy position from a Microsoft executive, not a settled international agreement.

Products worth knowing

Kudos Smart Wallet

An AI assistant that helps manage everyday money tasks.

Kudos is expanding its smart-wallet experience with tools that recommend which card to use for a purchase, activate eligible card offers, negotiate lower bills and identify forgotten subscriptions. Its latest version organizes these actions around a dashboard of money saved and potential next steps. The company says users keep all their savings rather than paying a cut to Kudos, and that the assistant does not move money without the user’s approval.

Some savings figures on the product page come from company-reported results and should be treated as estimates rather than guaranteed outcomes.

Worth a look if: you want help keeping track of card benefits and recurring bills, and are comfortable reviewing an app’s permissions and financial-data practices before connecting accounts.

Manifold MCP

Give your AI agent live marketing data.

Manifold MCP connects AI tools to information across marketing channels through a hosted Model Context Protocol server. It is designed to work with tools including Claude, ChatGPT, Cursor and Claude Code. Instead of asking an agent for generic growth advice, users can ask questions grounded in live data, such as which search terms are close to reaching page one, whether AI assistants recommend their company, what competitors are advertising or which relevant discussions are happening on Reddit.

The product says it is free to start with 500 credits, with paid usage after that. Its maker also provides a free library of marketing skills.

Worth a look if: you run a website or small business and want an AI assistant to work with actual SEO, competitor and audience data rather than relying on general marketing advice.

Ancher 2.0

Turn everything you read into useful work.

Ancher is an AI knowledge workspace that collects articles, videos, PDFs, posts and ideas in one place. Instead of manually organizing everything into folders, users can search across their saved material, ask questions about it and connect related sources. Its latest version focuses on turning saved knowledge into practical outputs such as research briefs, reports, presentations and content written in the user’s own voice. Ancher also offers a monitoring feature that can alert users when a relevant development or breakthrough appears.

Worth a look if: you regularly research topics, save more information than you can revisit, or want to turn scattered reading into reports and useful content.

Everything else in AI

Ultrahuman wants to estimate biological age across the UAE

Smart-ring maker Ultrahuman has signed a two-year memorandum of understanding with the Abu Dhabi Public Health Center to develop and validate a biological-age model. The system is intended to combine clinical biomarkers with wearable data, including signals related to sleep, activity, cardiovascular health and metabolism.

The goal is to estimate how a person’s biological markers compare with expected patterns for their chronological age and identify factors that may be associated with faster or slower ageing. The company says the model is designed to work with different types of wearables, not just its own smart ring.

The project will spend the next two years on research and clinical validation. A wider rollout is planned in phases, but no specific timeline has been announced. Ultrahuman says the model will be applied to UAE data rather than trained on it, and that it will not have access to personally identifiable information through the programme.

Wearable companies are moving beyond selling devices that track sleep and activity. They want to turn that data into health insights that could eventually be used at a population level. The important next step is validating whether those estimates are reliable and useful for real health decisions.

Worth your weekend

→ Read our blog: Notion AI vs Otter vs Fireflies for Meeting Notes 2026

→ View New Tool on Stackari

→ Missed a note? Browse the archive

Want a note in your inbox?

Join Noted for 5–7 things worth understanding in AI, delivered daily.

Free, daily, unsubscribe anytime.

See you tomorrow
same time, different note.

~ the Stackari team

P.S. Thanks for reading Noted. It means a lot. If you found this useful, pass it along to someone curious:)

Want to get in front of AI-curious founders?
Sponsor a note →

Was this note forwarded to you? Get your own →

Stackari · More news