In the Age of Agents, Do Knowledge-Centric Products Still Have a Future?

When Agents Learn to Work, Does Knowledge Accumulation Still Matter?

In 2025, AI Agent has become the hottest keyword in the tech world. From writing code automatically to autonomously completing complex workflows, Agents are taking over human execution-layer tasks at an astonishing pace. A pointed question has surfaced: when AI can already "get things done" on its own, do we still need a dedicated product for accumulating and managing knowledge?

Tencent ima's recently launched copilot mode provides an excellent case study for this debate. Positioned as a "knowledge brain," the product is trying to carve out an irreplaceable role amid the Agent wave.

Starting with a Lobster: ima Copilot's Real-World Performance

In the latest user review, someone used a vivid metaphor to describe the ima copilot experience — "from lobster to knowledge brain." Behind this metaphor was a complete end-to-end test, from a specific question to systematic knowledge construction.

The user tried to use ima copilot to complete a seemingly simple task: researching lobster cooking methods. A traditional AI chat tool would directly provide a recipe. An Agent-type product might automatically search, compare, and generate a complete cooking guide. But ima copilot works differently — it not only answered the immediate question but also automatically archived the relevant knowledge into the user's personal knowledge base, forming traceable, interconnected, and reusable knowledge nodes.

This difference may seem minor, but it points to a fundamental divergence: Agents solve "present problems," while knowledge accumulation solves "future problems."

Execution vs. Memory: The Essential Difference Between Two AI Paradigms

To understand the value of knowledge-centric products, we need to first clarify the two main trajectories of current AI products.

The first is the Agent paradigm. Its core logic is "give me a goal, and I'll handle it." Whether it's Manus, OpenAI's Deep Research, or various AutoGPT-type projects, they all fundamentally pursue stronger task execution capabilities. Users issue commands; the Agent breaks down tasks, calls tools, and outputs results. The entire process is efficient and direct.

The second is the knowledge management paradigm. Its core logic is "help you organize what you know and retrieve it when needed." Products like Tencent ima, Notion AI, and Mem follow this path. They don't treat single-task completion as the finish line but aim for long-term knowledge accumulation and structuring.

The difference between the two is like having an on-demand food delivery rider versus a well-maintained private kitchen. The rider is highly efficient, but you have to place a new order every time. A private kitchen requires upfront investment, but once built, you own a sustainable system of capability.

Three Key Design Decisions in ima's Copilot Mode

Based on the latest review, ima's copilot mode has made differentiated design choices on three levels:

1. Conversation as Accumulation

The biggest pain point of traditional AI conversations is their ephemeral nature — you spend an hour in a deep discussion with ChatGPT, but once you close the window, that knowledge is scattered across chat history and nearly impossible to reuse effectively. ima copilot addresses precisely this problem: every conversation can be automatically or manually converted into knowledge cards and incorporated into the user's personal knowledge system.

2. Knowledge Association and Rediscovery

Once knowledge accumulates to a certain scale, value lies not only in individual knowledge points but also in the connections between them. When users pose new questions, ima copilot can automatically link previously accumulated relevant knowledge, creating a positive feedback loop where the system "gets smarter the more you use it." This capability is currently very difficult for pure Agent-type products to provide — because Agents typically lack a persistent user knowledge graph.

3. A Closed Loop from Consumption to Production

ima copilot doesn't just help users consume information; it also supports content production based on existing knowledge bases. For example, after a user has accumulated extensive industry research notes, they can directly ask the copilot to generate analytical reports based on this private knowledge. This "accumulate first, produce later" model is essentially building the user's "second brain."

The Agent's Achilles' Heel: An Executor Without Memory

Returning to the question posed at the beginning: in the age of Agents, are knowledge-centric products truly irreplaceable?

The answer lies in the Agent's biggest current shortcoming — the lack of memory and context.

The vast majority of today's Agent products are "stateless." Every time a new task is initiated, the Agent starts from scratch. It doesn't know what you researched last week, doesn't understand the specific context of your industry, and cannot offer personalized advice based on three months of your thought processes. Even though some Agent products have begun introducing memory features (such as ChatGPT's Memory), they are far from achieving the depth of systematic knowledge management.

This is precisely the core moat of knowledge-centric products. When an Agent needs to understand "who you are," "what you know," and "what you care about," it must rely on a structured knowledge foundation. Without this foundation, no matter how strong an Agent's execution capabilities are, it remains just a worker without memory.

The Future: Convergence of Agents and Knowledge Accumulation

In truth, framing Agents and knowledge accumulation as opposing forces is a false dichotomy. The more logical evolutionary direction is deep integration of both.

In the short term, products like ima copilot need to prove the practical utility of knowledge accumulation — not collecting for collection's sake, but ensuring that accumulated knowledge genuinely contributes to subsequent decision-making and creation. If the knowledge base devolves into a "digital hoarder's" stockpile, product value will be significantly diminished.

In the medium term, knowledge-centric products will inevitably incorporate stronger Agent capabilities. ima's copilot mode itself exemplifies this trend — it is no longer just a passive knowledge base but an intelligent partner that can proactively help users complete tasks. Conversely, Agent products will also increasingly prioritize building a knowledge management layer, because an Agent without a knowledge foundation has an obvious ceiling.

In the long term, the ultimate product form will likely be an "Agent with memory" — one that can both execute tasks and remember everything about you, providing increasingly precise services based on that accumulation. Whoever completes this closed loop first will seize the advantage in the next generation of AI product competition.

Conclusion

In an era where everyone is talking about Agents, Tencent ima's decision to hold its ground in knowledge accumulation while evolving toward a copilot mode is a product strategy worth watching. It reminds us: the ultimate value of AI lies not in how many tasks it completes, but in whether it truly understands the person using it. And the prerequisite for understanding is memory; the vehicle for memory is knowledge.

While Agents compete over who can work faster, the product that quietly remembers everything for you may turn out to be the ultimate winner.