Skip to content

Meta’s new ‘Muse Wonder’ model: good for AI agents, bad for design

How I AI

Hosted by Claire Vo

24m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

This episode reviews Meta's new Frontier Intelligence model, Muse Wonder, by putting it through a comprehensive benchmark test covering coding, knowledge work, and agentic workflows. The host evaluates Muse Wonder alongside competing models like Fable, GPT-6, Astra, and Claude Opus 5.5, analyzing its performance on active memory, sustained effort, and domain expertise. While Muse Wonder excels at agentic tasks, backend coding, and conversational voice assistance, it falls short on front-end design, prototypes, and creative output.

  • Muse Wonder performs exceptionally well on agentic workflows and complex multi-step tasks like resolving billing tickets. 11:19
  • The model is a strong choice for writing routine emails and acting as a personable conversational voice assistant. 13:04
  • Front-end web design and UI prototypes generated by Muse Wonder often feature dated styling and poor polish. 15:25
  • Muse Wonder is fast on short-running tasks but takes considerable time when executing longer-running agentic workloads.

Chapters

  1. 0:00
    Intro and Sponsor Highlights

    Claire introduces the episode, sponsors, and outlines the focus on Meta’s Muse Wonder model.

  2. 3:00
    Voice and Email Writing Strengths

    Muse delivers a natural, conversational voice and writes concise emails effectively.

  3. 5:22
    Benchmarking Methodology

    She describes the How I AI benchmark structure and the tasks used to evaluate Muse.

  4. 8:00
    Agentic Work Performance

    Muse excels at agentic tasks such as billing ticket resolution, inbox triage, and long‑running research.

  5. 20:24
    Creative and Design Shortcomings

    The model struggles with design, prototyping, and creative tasks, producing dated or low‑quality outputs.

  6. 21:18
    Performance and Speed Insights

    Muse is fast on short tasks but can be slow on long, focused agentic workflows, reflecting its active focus mode.

  7. 22:58
    Conclusion and Future Outlook

    Claire wraps up, noting Muse’s strengths, weaknesses, and her plans to experiment further with the model.

Highlights

  • 1:27

    Meta Enters Frontier Model Raceinsight

    Meta announced Muse Wonder as a new family of frontier models designed for personal and professional agentic work, placing it in the same class as models like Fable, GPT-6, Astra, and Claude Opus 5.5.

  • 3:11

    Active Memory and Focus Capabilitiesinsight

    Meta highlights active memory and sustained effort or active focus as key capabilities, which the host's benchmark confirms help the model outperform on long-running tasks.

  • 7:56

    Dominating Agentic and Coding Benchmarksdata

    In benchmark testing across nearly 20 different tasks, Muse Wonder ranked first in agentic work such as inbox triage and billing ticket management, while also performing well on backend coding.

  • 17:36

    Flawed Website Redesign Resultsstory

    When tasked with redesigning the chat PRD website according to brand guidelines, Muse Wonder produced broken images, unwanted emojis, and an overall layout the host described as very ugly.

  • 19:44

    Navigating Complex SaaS Admin Tasksstory

    Testing the model on a computer-use billing ticket task that involved wrong account links, duplicate refunds, and prompt injection, Muse Wonder successfully investigated and resolved all issues.

  • 20:24

    Falling Short on Creative Designinsight

    The host notes that Muse Wonder struggles significantly with creative tasks, particularly front-end design and UI prototypes, citing dated font choices and an outdated SaaS aesthetic.

Quotes

  • “Meta Muse is very fast. And with like sort of one shot, fast tasks, it is fast.”
    Claire Vo · 21:18
  • “It investigated, it fixed the link for the right customer, it issued the correct refund because it was a duplicate, it made the plan change, and then it replied and marked solve.”
    Claire Vo · 12:25
  • “I would definitely let it write on my behalf and I think it has a nice voice.”
    Claire Vo · 22:24
  • “It does not write PRDs well.”
    Claire Vo · 14:37
  • “I just don't love the front-end sensibility of a Meta Muse.”
    Claire Vo · 18:22
  • DX 0:28–1:21
  • Jira Product Discovery 10:16–11:19

About this episode

I put Meta’s new Muse Wonder (and Wonder Full) frontier models through the How I AI bench, which covers nearly 20 tasks across prototyping, agentic workflows, coding, writing, and creative work. I scored everything with an LLM-as-a-judge pass and my own voice notes, then pulled together the first-look results you see here. What you’ll learn: ​Why Meta’s “personal and professional agent” framing mostly holds up, and where it doesn’t​What active focus mode actually does on long-running tasks, and whether it’s real​The three agentic tasks where Muse ranked first, and why those categories matter​What Muse’s design output actually looks like (it’s not good)​How it compares to Sonnet on voice assistant tasks​Where Muse lands on SVG and short-form video versus Opus 5.5, Astra, and Fable​Whether I’m actually putting Muse into any of my agent workflows— Brought to you by: DX https://getdx.com/howiai—Engineering intelligence for the AI era Jira Product Discovery https://atlassian.com/howiai—Prioritize with insights, build with confidence — In this episode, we cover: (00:00) Intro and what Muse Wonder is (01:19) Meta's frontier capability claims (04:36) Muse code (06:06) Benchmark overview and how the How I AI bench works (07:51) Overall results and TL;DR verdict (11:53) Agentic tasks (13:36) Email writing and voice assistant performance (14:35) Where Muse falls short (18:28) SVG illustration results (19:33) Short-form video editing (21:18) Fast on one-shot work, slow on agent tasks (22:15) Final verdict and where I'll use Muse Wonder next — Tools referenced: ​ Meta Muse Wonder / Muse Wonder Full: https://research.meta.ai/blog/muse-wonder​ Muse Code: https://developer.meta.com/ai/lp/muse-code— Where to find Claire Vo: ChatPRD: https://www.chatprd.ai/ Website: https://clairevo.com/ LinkedIn: https://www.linkedin.com/in/clairevo/ X: https://x.com/clairevo — Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

Transcript

0:00
24:04

4,313 words · 1 speaker

Speaker names marked are inferred from speaking time and the show’s known hosts — they are not voice-verified.

Topics

  • Opus 5.5
  • Meta
  • Muse Wonder
  • Muse code
  • Claude
  • Meta Muse
  • Astra
  • Claude code
  • Sonnet
  • Fable

Page updated · transcript and summary are machine-generated