19 February 2026

Meta’s AI Agent is Better Than OpenClaw (Manus AI Demo)

1Frameworks

8Insights

Frameworks in this episode

Innovation6 steps

Build-Test-Refine Skill Loop

Turn expert workflows into reusable AI skills through repeated testing and refinement.

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Hot Take· 3

Hot Take17:00

Personal Agents Will Become Teams of Specialists

The episode predicts a shift from one general personal agent toward multiple specialized agents with separate responsibilities. A user might talk to a content agent and an investment agent in different chats, while the agents exchange work with one another to complete broader workflows.

Current products commonly expose one general personal agent.
Future users may maintain separate agents for content, investing, and other domains.
Specialized agents could communicate and hand work to one another.
Messaging channels may become the interface for an entire virtual team.

“You're going to be able to create different agents who are specialized in different areas.”
17:00

“So, you can actually start to build teams of autonomous agents that can all talk to each other and you can talk to those.”
17:30

#multi-agent#specialization#future of work#agent teams

Hot Take20:00

Better Output Beats the Lock-In of AI Memory

The hosts argue that AI users will readily switch products when another system provides stronger output or a better interface. Even extensive stored memory may not create durable loyalty because competing tools also offer memory and users prioritize immediate value.

The hosts describe rapidly shifting their own usage toward Claude and Manus.
Convenient access to underlying models can matter more than using those models directly.
Stored memory proved less effective as a retention mechanism than expected.
Users are likely to gravitate toward whichever model or product currently performs best.

“I just think core value of the output is going to outweigh the stickiness of memory”
20:30

“I don't think AI users are loyal in any way. I think we will just gravitate towards whatever the best model to use is.”
21:00

#ai models#memory#user loyalty#product strategy

Hot Take21:30

The Skill Economy Makes Curation the New Bottleneck

With repositories already offering tens of thousands of agent skills, obtaining baseline expertise is becoming easier. The harder problem becomes choosing trustworthy skills, integrating them into real workflows, and adapting generic instructions to proprietary needs. Exceptional practitioners may gain leverage by encoding distinctive expertise into higher-quality skills.

One featured marketplace reportedly contains 47,000 skills.
Abundant skills make selection and integration harder than basic acquisition.
Generic skills still need adaptation to the user's context and workflow.
Top practitioners can multiply distinctive expertise by turning it into reusable systems.

“The challenge is not how do I build skills or get them? It's going to be how do I choose the right ones and integrate…”
22:00

“if you are 10x better than everybody else at something, you can now make a skill around that and get 10x more output at a…”
22:00

#skill marketplace#expertise#curation#leverage#future of work

Explainer· 2

Explainer00:30

Why Manus Is an Easier Starting Point Than OpenClaw

The hosts position Manus as a managed personal autonomous agent that offers much of OpenClaw's utility without requiring users to configure their own computer or cloud infrastructure. Telegram access, memory, scheduled jobs, and reduced setup complexity make it more approachable for nontechnical users, although the discussion does not establish that it eliminates every possible security risk.

OpenClaw offers strong autonomy but requires more technical setup.
The hosts highlight security concerns surrounding self-configured agent systems.
Manus provides Telegram access without requiring a Mac mini or AWS deployment.
Memory and scheduled jobs support recurring personal workflows.

“The challenge is it's like kind of complex to set up. There's a bunch of security risks. Like, you need to be a pretty technical…”
00:30

“you don't need to write code, you don't need a Mac mini or AWS or any of these things to run it.”
03:00

#manus#openclaw#ai agents#security#automation

Explainer07:30

Skill Files Make Agent Expertise Portable

The hosts explain that skill.md files are not proprietary to Manus and can move among compatible agent products. This portability reduces dependence on one interface because users can carry their accumulated workflow instructions from Claude or ChatGPT into Manus and other systems.

Skills are stored as markdown instruction files.
The same skill can be used across multiple compatible agent systems.
Portable skills preserve workflow knowledge when users change products.
The hosts distinguish a reusable system skill from a one-off conversational prompt.

“And so, skills is basically a system, whereas prompt is conversation.”
07:30

“If you create a skill.md file, which is how I teach the AI how to do something, I can use it in all of those…”
09:00

#skills#portability#markdown#ai agents

Story· 1

Story04:30

A Dog-Walk Voice Note Produced a Finished Research Brief

One host sent the agent a voice note asking it to collect high-performing Reddit threads and articles about OpenAI's rumored OpenClaw acquisition. By the time he returned from walking his dog, the agent had ranked the research, extracted content angles, and presented the findings in an interactive page tailored to his audience.

The assignment was initiated entirely through a phone voice note.
The agent gathered and ranked recent sources by engagement.
It converted research into educational and provocative content angles.
The output reflected an audience of CMOs, founders, and marketing executives.

“Just sent it a voice message. That is the kind of unlock with autonomous agents, right? All I did was send it a voice note.”
04:30

“You can get a great first pass when you are on your phone doing any other thing, and you just tell your agent to create…”
06:30

#voice notes#research#content marketing#autonomous agents

Tool· 1

Tool15:30

Forward an Email and Let the Agent Run a Workflow

Manus provides an email address that can receive forwarded messages and trigger predefined workflows. The hosts describe using it to research investment opportunities, analyze newsletter claims, draft replies, and create personalized summaries with relevant action items.

The personal agent has its own email inbox.
Forwarded deal flow can trigger company research and bull-versus-bear analysis.
Newsletters can be converted into personalized summaries and action items.
Other people can be copied into the resulting email thread for collaboration.

“It has an inbox. So, basically, you have this email address and you can just email your Manus bot.”
15:30

“whenever deal flow comes in, I'm going to say, "Research the company looking for investment and give me the uh, bull and bear case."”
15:30

#email#workflows#deal flow#summarization#collaboration

Takeaway· 1

Takeaway18:30

Use Token-Hungry Agents Only for Valuable Work

The convenience of a super-agent comes with meaningful usage costs, especially when workflows generate images or access premium data. One host reports consuming roughly 20,000 credits while developing an advertising workflow and recommends reserving heavy agent use for strategically worthwhile tasks.

Image generation and premium data access can accelerate credit consumption.
Iterative development may be substantially more expensive than simple requests.
A super-agent hides the complexity of coordinating multiple models but not their cost.
Low-value email interpretation is a poor use of expensive autonomous processing.

“I burned through probably 20,000 credits just iterating through my ad optimizer trying to get that right.”
19:00

“Yes. >> so, you do want to use it for you know, strategic worthy things.”
19:30

#ai costs#credits#task selection#manus