Building Personal Agents, Not General Agents?

We are living through a true technological transformation. In the internet era, every pivotal wave, from search engines that organized scattered information, to mobile devices that untethered access from desktops, to cloud computing that democratized scalable storage and processing, and to recommendation systems that personalized how we discover content, has been built on one deceptively simple foundation: the hyperlink. Information, media, and services became connected, and with each cycle, productivity and participation expanded.

Yet the revolution brought by AI is breaking beyond the framework of “connection.” What excites me most is not that it connects and organizes hyperlink resources beyond the internet, but that it can deliver results with everyone’s participation and do so equally for everyone. It feels early, fragile, and full of opportunity. For those who want to build, imagine, and create, it is a rare moment.

At Meland Labs, we have always believed: the ultimate value of AI does not lie in building abstract technical platforms, but in rooting itself in the real needs of individuals, protecting your attention from being diluted by the information flood, empowering your growth to overcome inefficient processes, and unlocking your value from being trapped by tool barriers, etc. In short, it makes your work and life easier, helps you grow, and builds your wealth. It is for this reason that we clarified our direction from the very beginning: focus on personal advanced AI agents, rather than chasing the illusion of general agents.

Why “Personal Agents” instead of “General Agents”?

A general agent may function as a basic “virtual machine based task execution platform” with cross-scenario fundamental capabilities, but AI that truly solves individual pain points must be “targeted”:

• For professionals, what is needed is an agent that accurately understands their long-term work context, it knows your client communication style and project progress rhythm, efficiently organizes meeting minutes and follows up on tasks, rather than a general tool that “knows a little about everything but masters nothing” across all industries.

• For learners, what is needed is an agent that adapts to personal learning habits, it remembers the knowledge points you struggle with and your preferred learning methods, generates review outlines and recommends supplementary materials on demand, rather than a “knowledge database” that only outputs standardized content.

• For daily life scenarios, what is needed is an agent that aligns with your personal lifestyle, it understands your dietary preferences and travel habits, intelligently plans itineraries and reminds you of important schedules, rather than a “general assistant” that recommends the same solutions to everyone.

In fact, compared to general-purpose agents, advanced personal agents face a very different set of technical challenges and priorities. They must ensure absolute privacy and security, build deep user modeling through long-term habit learning, and maintain ultra-long-term memory and planning abilities. They need to serve proactively rather than passively, and most importantly, they must internalize and accurately process large volumes of highly private, multimodal data. This is not just an incremental improvement over a “universal automation agent”, it is a fundamentally different design philosophy. While general agents often resemble task platforms built on virtual machines and automation pipelines, personal agents must be closer to companions: secure, contextual, and uniquely tailored to you.

Complex fields require professional depth, and individual needs demand personalized adaptation. General agents pursue “breadth” but struggle to achieve “depth”; by contrast, personal agents center on the “individual”, they can take root in specific scenarios while continuously accumulating interaction context with you, ultimately delivering a service experience that is “more understanding of you than you are of yourself.”

What we’ve chosen is to build ‘bundled’ and ‘depth’ AI agents, assistants that operate in a unified, AI-native way, rather than forcing you to piece together endless software. Our goal is simple: to serve every person equally, even if it means tackling the difficult, messy work behind the scenes. At Meland Labs, every small personal scenario matters.

Building Truly Valuable Personal Agents Requires Breaking Through Three Core Pillars

To turn personal agents from a “concept” into a “necessity,” we must move beyond mere “feature stacking” and build a “result-oriented” service system centered on individual needs. At Meland Labs, we break this down into three key directions:

1. Long-term context and tools: AI needs access to the full, relevant context from your real scenarios, paired with an open ecosystem of tools.
2. Results that accumulate over time, not functions: Instead of navigating menus and features, you should simply get outcomes, just as a human helper would deliver.
3. Communicative, composable interfaces: A new interaction model, not typical webpages or documents, but something more fluid: conversations, result spaces, and generative apps, seamlessly woven together by AI.

To get there, we must tackle two hard problems, harnessing the full power of AI:

1. How to record, reuse, and share intermediate context and results over long and fragmented time cycles, not just temporary or one-off answers.
2. How to bring multiple threads of interaction together, adjusting, merging, and moving them forward, while enabling greater human collaboration along the way.

If we solve these, AI products will reach the same robustness and practicality people expect from internet products today.

We don’t believe AI will replace people. Instead, it will replace the glue work that forces humans to jump between tools and processes. The person remains the brain, the driver. AI simply accelerates the journey from intention to outcome. In this sense, I envision a future where AI-native products gradually replace traditional software, forming a human-centered operating system for both work and life. We also don’t believe a ‘general agent’ is a magic solution. Complex domains demand expertise, not one-size-fits-all answers.

Our First Step: Starting with “Protecting Attention” to Launch Lumis

Based on our thinking about personal agents, our first product, Lumis, a brand new “communication agent”, a choice rooted in our team’s own pain points: today’s work is already overwhelmed by massive information: unread messages on multiple platforms, pending emails in your inbox, and @mentions in collaboration software. Every day, just “processing information” consumes a huge amount of energy, leaving little time for tasks that truly require thinking.

We realized that for individuals, attention is the most scarce resource; for personal agents, protecting your attention is the most fundamental and important value.

Therefore, Lumis does not pursue “all-purpose functions” but focuses on “attention management in communication scenarios”: it can move between various conversations just like you, automatically summarizing lengthy chats over time to extract key information; drafting replies that match your tone to avoid “template-style responses”; and even marking “urgent matters” and “important needs” to help you filter out useless information.

Its goal is not to “replace you in communication,” but to save you from the tedious processes of “information screening, content summarization, and initial replies,” freeing up precious mental space for you to focus on truly important creation and decision-making.

Currently, Lumis is in its beta testing phase. It is not perfect: it may occasionally misunderstand your tone preferences and needs further adjustments in complex communication scenarios. However, every user who joins the test is helping us refine it, your every piece of feedback contributes to our answer to “how personal agents can better serve individuals.” For a small team like ours that focuses on “small scenarios, big value,” these needs from real individuals are more valuable than any technical parameters.

Final Note: Meland’s Original Intention

Why ‘Meland Labs’? The name comes from my daughter. There’s a local playground she loves called Meland. Watching the children there, collaborating, experimenting, and losing themselves in play, I was struck by how focused and alive they seemed, as if time itself slowed down. That’s the feeling I want Meland Labs to create for everyone: a place where, even in a chaotic world, you can regain focus and make time feel abundant again.

This is exactly the experience we hope to bring to everyone through personal agents: in an era of information explosion and fast-paced life, let AI take on those trivial “connecting tasks” for you, help you reclaim your fragmented attention, and allow you to refocus on what you truly love, whether it’s creation at work or companionship in life, with the focus and ease of “time slowing down.”

Meland Labs is just beginning. Our ambition isn’t to chase hype or promise magic, but to build patiently, person by person, assistant by assistant, toward a future where AI quietly empowers you to live, work, and grow with greater freedom.

And perhaps, in the middle of all the noise, to help you feel, even if only for a moment, that time has slowed down again.

Ethan

Founder, Meland Labs