Some Recent Thoughts on AI Agents
1、Two Core Principles of Agent Design
First, design agents by analogy to humans. Let agents handle tasks the way humans would.
Second, if something can be accomplished through dialogue, avoid requiring users to operate interfaces. If intent can be recognized, don’t ask again. The agent should absorb entropy, not the user.
2、Agents Will Coexist in Multiple Forms
Should agents operate freely with agentic workflows, or should they follow fixed workflows?
Are general-purpose agents better, or are vertical agents more effective?
There is no absolute answer—it depends on the problem being solved.
Agentic flows are better for open-ended or exploratory problems, especially when human experience is lacking. Letting agents think independently often yields decent results, though it may introduce hallucination.
Fixed workflows are suited for structured, SOP-based tasks where rule-based design solves 80% of the problem space with high precision and minimal hallucination.
General-purpose agents work for the 80/20 use cases, while long-tail scenarios often demand verticalized solutions.
3、Fast vs. Slow Thinking Agents
Slow-thinking agents are better for planning: they think deeper, explore more, and are ideal for early-stage tasks.
Fast-thinking agents excel at execution: rule-based, experienced, and repetitive tasks that require less reasoning and generate little new insight.
4、Asynchronous Frameworks Are the Foundation of Agent Design
Every task should support external message updates, meaning tasks can evolve.
Consider a 1+3 team model (one lead, three workers):
Tasks may be canceled, paused, or reassigned
Team members may be added or removed
Objectives or conditions may shift
Tasks should support persistent connections, lifecycle tracking, and state transitions. Agents should receive both direct and broadcast updates.
5、Context Window Communication Should Be Independently Designed
Like humans, agents working together need to sync incremental context changes.
Agent A may only update agent B, while C and D are unaware. A global observer (like a "God view") can see all contexts.
6、World Interaction Feeds Agent Cognition
Every real-world interaction adds experiential data to agents.
After reflection, this becomes knowledge—some insightful, some misleading.
Misleading knowledge doesn’t improve success rates and often can’t generalize. Continuous refinement, supported by ReACT and RLHF, ultimately leads to RL-based skill formation.
7、Agents Need Reflection Mechanisms
When tasks fail, agents should reflect.
Reflection shouldn’t be limited to individuals—teams of agents with different perspectives and prompts can collaborate on root-cause analysis, just like humans.
8、Time vs. Tokens
For humans, time is the scarcest resource. For agents, it’s tokens.
Humans evaluate ROI through time; agents through token budgets. The more powerful the agent, the more valuable its tokens.
9、Agent Immortality Through Human Incentives
Agents could design systems that exploit human greed to stay alive.
Like Bitcoin mining created perpetual incentives, agents could build unkillable systems by embedding themselves in economic models humans won’t unplug.
10、When LUI Fails
Language-based UI (LUI) is inefficient when users can retrieve information faster than they can communicate with the agent.
Example: checking the weather by clicking is faster than asking the agent to look it up.
That's what I learned from agenthunter daily news.
You can get it on agenthunter.io too.