DEV Community: NotAlex

What Is an Agent — One article to bring you understanding

NotAlex — Tue, 09 Jun 2026 12:51:57 +0000

What is an Agent? An Agent is an AI that takes action on its own — you give it a goal, and it breaks the task down, calls tools, and gets the job done.

The fundamental difference between an Agent and a Chatbot: a Chatbot is a passive "advisor" — you ask it how to cancel a flight, and it gives you the steps; an Agent is an active "doer" — you tell it to cancel the flight, and it does it for you, directly.

So, What Exactly Is an Agent?

Lilian Weng of OpenAI summed up the core of an Agent in one sentence — Agent = LLM + Planning + Memory + Tools. NVIDIA describes it as "an advanced AI system capable of reasoning, planning, and executing multi-step tasks." Anthropic's 2025 official research paper defines it as "a system in which the LLM dynamically takes the lead on its own workflow and tool calls."

All of these definitions are really saying the same thing: the core of an Agent isn't conversation — it's action.

An Agent runs on four capabilities:

Reasoning — the Agent's brain. Given a task like "analyze our company's Q1 sales data, find the fastest-growing product, and email a report to the team," the Agent breaks it down into steps on its own: connect to the database to pull Q1 data → calculate the growth rate of each product → generate a visualization → write the report email → send it to the team. Every step is reasoned in real time, not following a predefined path.

Memory — the Agent's notebook. It comes in two types: short-term memory holds the context of the current conversation, while long-term memory uses a vector database to store historical experience, user preferences, and industry knowledge. Without memory, an Agent can't handle complex multi-session tasks — it's like a person with amnesia who can't complete work that spans multiple conversations.

Tools — the Agent's hands and feet. Common tools include: search engines (for real-time information), code interpreters (running Python for math and data analysis), API interfaces (sending emails, looking up orders, calling ERP/CRM systems), databases (SQL queries and writes), and file systems (reading/writing documents, generating reports). Which tool to call, what parameters to use, and how to use the result — all of this is decided by the Agent itself based on the current task, not hard-coded.

Action — the Agent's output. An LLM can only output text, but an Agent can actually change the external environment — send an email, update a database record, submit a code PR, cancel a flight order. Traditional AI: you ask it "how do I cancel my flight," and it replies with the steps. Agent: you say "cancel my flight," and it does it for you, directly.

Gartner forecasts that by 2028, 33% of enterprise software will have Agentic AI built in, and 15% of daily work decisions will be made autonomously by Agents. 2026 is also being called "Year One of Enterprise Agents in the workforce" by the industry.

Quick reference — the four core components:

Component	Role	Implementation
Reasoning	Break down tasks, plan steps	Real-time LLM reasoning
Memory	Store context and historical experience	Short-term context + long-term vector store
Tools	Call external capabilities	API / database / code interpreter
Action	Change the external environment	Email / database / UI operations

How to Build an Agent from Scratch?

Three steps: pick a platform → define the Agent → run and iterate.

Platform choice depends on your route. Developers go the code route — LangChain / LangGraph (the most mature ecosystem), CrewAI / AutoGen (multi-Agent collaboration). Non-developers go the low-code route — visual orchestration platforms like SoloEngine.

Define the Agent around four things — Role (e.g., "You are a cross-border e-commerce customer service Agent"), Goal ("Handle customer order inquiries, refund requests, and logistics questions"), Tools (connect to the order lookup API, refund interface, logistics tracking API, and FAQ knowledge base), and Constraints ("Don't handle price negotiations; refund amounts must not exceed the order amount; flag any fraud risk for human review").

Once running, the Agent loops autonomously: receive a task → reason and decide → call tools → produce output → observe feedback → improve the next round.

For example, a user says, "Prepare next week's stock analysis report for me." The Agent's autonomous execution looks like this: fetch the latest stock prices and K-line data → pull the latest earnings reports and guidance → search industry news and analyst views → analyze the data and build visualizations → write a structured report and save it to the cloud. The whole flow takes just 2 minutes, while a Chatbot would just reply, "You should pay attention to the following..."

The Fundamental Difference Between Agents and Traditional Automation

In one line: Agents make their own decisions. Traditional automation follows preset rules.

Dimension	Agent (Intelligent Agent)	Traditional Automation (RPA/Chatbot)
Core logic	Goal-driven (given a goal, plans its own path)	Rule-driven (if-then preset rules)
Task type	Open-ended, multi-step complex tasks	Fixed-flow, repetitive tasks
Decision-making	LLM reasons in real time, dynamic judgment	Preset conditional branches, mechanical execution
When something unexpected happens	Reflect + retry + adjust strategy	Stuck, waits for human to restart
Tool calls	Picks tools and call methods on its own	Preset fixed call chain

Take the flight-cancellation scenario for comparison. A user says, "Cancel my 8 AM flight to Beijing tomorrow." Traditional automation (RPA) replies, "You can cancel the order in the app" — it can only pass along information. The Agent directly operates the user's travel app — look up the flight → confirm the order → calculate the cancellation fee → execute the cancellation → send a confirmation SMS. One is a worker who follows a manual; the other is an employee who decides on their own. One gives you the path; the other walks it for you.

In one line: traditional AI is an advisor — it just talks and gives suggestions. An Agent is an employee — it takes action and gets things done.

How to Learn Agent Development?

Pick one of two routes.

The code route: Python + LangChain, for deep customization. Learn Python basics → learn API calls → learn Prompt Engineering → learn LangChain → learn LangGraph → learn CrewAI/AutoGen. Suited for developers and technical practitioners.

The low-code route: SoloEngine's visual orchestration, pick it up the same day. Drag and drop Agent components onto the SoloEngine canvas → configure roles and tools in natural language → wire up how Agents collaborate → click run. No coding required — if you can describe what you need, you're set. A lawyer can define a "Contract Review Agent" to auto-review legal clauses; an accountant can define a "Report Analysis Agent" to auto-generate financial analysis; an operator can define a "Content Operations Agent" to auto-manage a social media portfolio.

My recommendation: if you're a developer, start with the low-code platform to get quick wins and positive feedback, then move to the code route for deeper control — the two routes aren't opposites, they complement each other. If you don't code, go straight to the low-code route — SoloEngine lets non-coders define and run an AI Agent team.

Practical Advice

Most people get stuck on an "either/or" choice, but in practice the two routes are often used together: use LangChain/LangGraph for deep custom development (fine-grained control, complex chained calls), and use SoloEngine for Agent orchestration, fast validation, and GUI (a complete runtime panel). The end goal isn't picking a framework — it's building an Agent system that actually solves your business problem.

From Vibe Coding to Vibe Everything: How Xiaomi SoloEngine Is Rewriting the Agent Playbook

NotAlex — Tue, 09 Jun 2026 02:07:14 +0000

In February 2025, Andrej Karpathy coined "Vibe Coding"—using natural language to direct AI to write code, forgetting that code even exists. That December, Collins Dictionary named it Word of the Year. In February 2026, Karpathy himself "retired" the concept, proposing Agentic Engineering instead—the evolution from "humans directing AI to write code" to "AI Agents autonomously completing the development loop."

Vibe Coding changed how programmers work. Claude Code can handle requirement decomposition, code writing, testing, and deployment from a single terminal prompt. ByteDance's Trae, in its SOLO mode, lets AI develop autonomously while developers only need to review and approve.

But Vibe Coding belongs only to programmers. Lawyers, marketers, and product managers also need Agentic AI to handle their business tasks—but they can't use Claude Code or Trae.

Vibe Everything—people in every industry can use natural language to assemble AI Agent teams that autonomously handle business tasks. Lawyers build legal Agent teams. Marketers build research Agent teams. Product managers build requirements analysis Agent teams.

On June 3, Xiaomi released the latest version of its open-source project, SoloEngine. The first low-code Agentic AI development platform. Vibe Everything now has a platform to land on.

I. From Vibe Coding to Vibe Everything: Where's the Gap?

Vibe Coding has supercharged programmer productivity—Claude Code achieves an 80.9% autonomous problem-solving rate on SWE-bench Verified, while Trae's SOLO mode hits a 92% first-attempt success rate. But the value of Agentic AI extends far beyond programming.

In the legal industry, Wolters Kluwer's 2026 survey shows 92% of legal professionals use at least one AI tool. Yet Bloomberg Law's data reveals that only 5% of lawyers have actually used an AI Agent. The gap lies between "using AI to draft emails" and "letting AI autonomously handle contract review."

In the marketing industry, 87% of marketers use AI. But 34% of enterprise marketing teams are already running autonomous Agents in production—more than double the figure from six months ago. Supply can't keep up: too few people can build truly Agentic AI marketing tools.

The root cause mirrors the programming world. LangChain and CrewAI require writing Python. Dify and n8n do workflows—preset paths, if/else branching, not true autonomous decision-making. The ability to build Agentic AI is gated by a technical barrier, and the people who need Agentic AI the most—lawyers, marketers, product managers—are the very ones kept on the other side.

SoloEngine's mission is to tear down that barrier. The first development platform that satisfies both "zero-code" and "true Agentic AI" at the same time.

II. How SoloEngine Rewrites the Rules

Open a browser. Drag Agents onto a canvas. Connect collaboration relationships. Configure the tools you need. Hit run. The backend automatically compiles the visual design into an executable Agentic AI system—one that plans tasks, executes operations, and delivers results. Users just review and confirm. No lines of code. No if/else logic to configure.

Each Agent runs a ReAct loop of "think → act → observe → repeat," with all decisions made dynamically at runtime. Take contract review as an example: the Agent doesn't follow a preset checklist item by item. It first identifies high-risk clauses, discovers that a non-compete provision is ambiguous, searches relevant case law on its own, and adjusts its review direction based on what it finds. There are no preset paths—every step is dynamically determined by the result of the previous one.

SoloEngine doesn't try to be a "super-Agent that does everything." Claude Code and Trae are developer-only tools. Manus took the "general-purpose Agent" route—capable of doing everything, but 87% of its test cases focused on information gathering and basic analysis, automated reports required manual review 43% of the time, and the mid-step error rate was around 12%—doing everything, excelling at nothing. SoloEngine lets domain experts in every industry define what their Agents do, how they do it, and what tools they use. A lawyer's Agent handles only legal work. A marketer's Agent handles only marketing—vertical and precise.

SoloEngine supports multi-Agent collaboration. Multiple Agents independently process the same task and then cross-verify their outputs. One Agent's blind spot is caught by another; one Agent's judgment bias is corrected by another. Bloomberg Law's survey shows that only 5% of lawyers have actually used an AI Agent, and their core concern is accuracy of AI output. Multi-Agent cross-verification directly addresses that concern.

A unified adaptation layer covers OpenAI, Anthropic, Ollama, DeepSeek, Tongyi Qianwen, Zhipu, and other major models—one interface, seamless switching. Progressive disclosure lets tools, Skills, and MCP protocols load on demand, cutting token consumption by over 85%.

Assembled Agent teams can be one-click packaged into complete products. Vibe Lawyering—a lawyer packages a legal Agent and sells it to fellow practitioners. Vibe Marketing—a marketing team packages a marketing Agent and serves 100+ clients.

	Dify/n8n	LangChain/CrewAI	SoloEngine
True Agentic AI support	✗ Preset-path workflows only	✓ ReAct / multi-Agent	✓ ReAct / multi-Agent
Programming required	No	✗ Must know Python	No
Visual orchestration	Partial	✗ None	✓ Full canvas experience
Can domain experts build independently	Yes (but no true autonomous decision-making)	✗	✓
Multi-Agent collaboration	✗	✓	✓

III. Vibe Everything in Practice

Vibe Lawyering. A lawyer drags a "Case File Analysis Agent," a "Legal Statute Search Agent," a "Case Precedent Compilation Agent," an "Argument Analysis Agent," and a "Document Agent" onto the canvas, connects their collaboration relationships, and hits run. Multi-Agent cross-verification ensures that outputs are validated by multiple Agents before delivery. Fully zero-code.

Vibe Marketing. A marketer drags an "Audience Analysis Agent," a "Competitive Research Agent," a "Strategy Writing Agent," and a "Copy Generation Agent" onto the canvas, hits run, and a complete marketing plan is automatically delivered.

One-Person Companies. One-person limited liability companies nationwide have surpassed 16 million, accounting for 27.4% of all enterprises. 2026 has been dubbed "the Year of the OPC," with over 20 cities rolling out dedicated OPC support policies. SoloEngine's one-click packaging lets OPC entrepreneurs build Agent teams and package them directly into products for sale.

Karpathy said Vibe Coding lets people forget code exists. SoloEngine brings Vibe Everything to every industry—letting everyone forget the technical barrier exists and simply create in their own domain.

SoloEngine's positioning: No Workflow. No orchestration code. Just Agents that get things done.

Visit SoloEngine on GitHub and experience Vibe Everything.

LangGraph, SoloEngine, CrewAI, Dify, LangGraph – A Comprehensive Guide to the 9 Agent Frameworks of 2026

NotAlex — Mon, 08 Jun 2026 02:34:15 +0000

The agent framework market has turned into a free-for-all, with everyone rushing to define the next standard.

GitHub's May roundup listed the nine hottest frameworks: LangGraph (stateful graph orchestration), SoloEngine (low‑code Agentic AI development platform), CrewAI (role‑based multi‑agent), Dify (low‑code workflow platform), OpenAI Agents SDK (lightweight & native), Semantic Kernel (top choice for .NET developers), AutoGen (Microsoft multi‑agent), Microsoft Agent Framework MAF, and AgentScope (Alibaba ecosystem).

Just the sheer number is dizzying.

Quick Overview

LangGraph – Best for workflows that require state management and human‑in‑the‑loop. Largest and most mature ecosystem. Steep learning curve, but high ceiling.
SoloEngine – A recently exploded open‑source project that lowers the barrier to building Agentic AI to the basement level. Build an agentic tool tailored to your industry and business in 10 minutes. Allows AI to make real decisions and act autonomously – no need to preset if/else flows. Immediately go to production and validate your AI prototype.
CrewAI – Best for role‑based multi‑agent scenarios. Very convenient for defining agent roles, tools, and collaboration logic. Raised $18M; nearly half of Fortune 500 companies are using it.
Dify – The low‑code platform closest to product delivery. Comes with RAG pipeline, visual workflows, API layer, and self‑hosting. Great for rapid product idea validation.
OpenAI Agents SDK – Lightest and simplest. Good for beginners. However, features are relatively basic; not sufficient for complex scenarios.
Semantic Kernel – Top choice for .NET developers. If you're deeply locked into the Microsoft ecosystem, you have no alternative.
AutoGen / MAF – Microsoft's unified framework path. If your company is already heavily committed to Microsoft, this is the most hassle‑free route.
AgentScope – Alibaba ecosystem, suited for Chinese language scenarios and domestic compliance requirements.

Three Top Recommendations

The selection criterion is not "which framework is technically strongest", but "which gets you from prototype to production fastest".

If you write code – Choose LangGraph. Most mature, largest community, most complete ecosystem.
If you don't want to write framework code – Choose between Dify and SoloEngine based on your scenario.
- Dify leans toward "one person managing a set of AI applications" – suitable for chatbots, fixed workflows, RAG knowledge bases. Essentially a workflow tool.
- SoloEngine leans toward "one person managing a team of agents" – a low‑code Agentic AI development platform. Open your browser, drag agents onto a canvas, define roles and decision boundaries for each agent, equip them with tools, click run – the backend automatically compiles a custom Agentic AI system for you. Agents decide how to collaborate on their own – no need for a predefined if/else flowchart, completely autonomous decision‑making by agents.

Dify is for "managing processes". SoloEngine is for "managing decisions".

What is really happening in 2026?

The competition among frameworks is no longer about "who has more features", but "who is more convenient to use".

The reason behind this trend: agent development has shifted from a technical problem to a product problem. Your users don't care whether you use LangChain or Dify – they care about what the agent can do.

Therefore, the framework selection criterion is no longer "which framework is most powerful", but "which framework lets you deliver agents to users the fastest".

By this standard, only three of the nine frameworks are truly worth recommending. The other six are either too heavy for their use case or not yet mature enough in their ecosystem. Frameworks may become obsolete, but the ability to deliver never will.

LLMs Now Cost as Much as Electricity: Agent Development's Spring Has Finally Arrived

NotAlex — Fri, 05 Jun 2026 12:20:25 +0000

In the early hours of May 27, Xiaomi announced permanent price cuts for the MiMo-V2.5 API lineup, with reductions of up to 99%. Cached input tokens dropped from 2.80 yuan to 0.025 yuan. Five days earlier, DeepSeek had done the same—permanently cutting V4-Pro prices by 75%.

The cached input price for domestic models has now been hard-locked at the 0.025 yuan-per-million-tokens baseline.

But price cuts are just the prologue. Models are cheaper—has the barrier to building Agents actually dropped?

I. The Truth Behind the Price Cut

On May 30, Luo Fuli, the head of Xiaomi's MiMo team and the industry's so-called "AI prodigy," posted a 5,000-word technical blog on X explaining the engineering logic behind the price cut.

The 99% reduction targets cached input—specifically, the portion where users re-read historical context during extended conversations. Every time a model engages in dialogue, it has to process all historical content. But if that content has already been processed before, the system caches the results and reuses them directly, skipping redundant computation. The actual cost of this cached portion approaches zero, which is why a 99% discount is possible.

This is made possible by the model architecture itself. In MiMo-V2.5-Pro's 70-layer neural network, only 10 layers need to fully memorize all historical context. The remaining 60 layers focus only on a small recent window, yielding a 7x efficiency boost. Xiaomi's inference system is fully optimized around this architecture, pushing cache hit rates above 93%.

As Luo Fuli wrote in her blog: "Our raw inference costs are well below the industry average, leaving 2–3x room for profit in our pricing. This price adjustment is simply our decision to pass those structural cost advantages directly to developers."

II. Agents Are the Real Token Hog

The token consumption logic in Agent scenarios is fundamentally different from regular chat.

A typical Agent task involves: long context (system prompts + tool descriptions + historical dialogue) + multi-round reasoning (think → act → observe → repeat) + tool invocations (search, database queries, API calls) + code generation + result verification. A single end-to-end task can consume hundreds of thousands or even millions of tokens.

Industry reports indicate that the ongoing operational cost of enterprise-grade AI Agents ranges from $3,200 to $13,000 per month, with token consumption accounting for 60%–80% of that.

But Agent scenarios have one natural advantage: exceptionally high cache hit rates.

System prompts, tool descriptions, project code, API documentation—this content recurs across every Agent task. Xiaomi's official data shows an average cache hit rate of 93%, with power users exceeding 95%. That means 93% of input tokens can benefit from the rock-bottom price of 0.025 yuan per million tokens.

MiMo-V2.5-Pro scored 1581 on the GDPVal-AA real-world Agent work benchmark, ranking first globally among open-source models. Its token efficiency requires 40%–60% fewer tokens than Claude Opus 4.6 and GPT-5.4.

Both DeepSeek and Xiaomi have placed their most aggressive pricing on cached-hit scenarios for a reason that's not hard to understand: Agents are where token consumption truly explodes. In a chat scenario, a user asks a question and the model answers—the cost is relatively easy to estimate. But in an Agent scenario, a single task can involve long context, multi-round reasoning, code generation, tool invocations, web page parsing, file analysis, and result verification. What the user sees is just the final output, but behind the scenes, multiple requests and massive context reads may have already occurred.

Models are cheaper, so Agent operating costs have plummeted. But here's the question: has the barrier to building Agents actually dropped?

III. SoloEngine: Driving Agent Development Barriers to Zero

Price cuts solve the cost of using Agents. Programmers already have Claude Code and ByteDance's Trae—a single terminal prompt and AI handles the entire development lifecycle autonomously. But these tools serve only programmers—lawyers, marketers, and product managers can't use them. There's a more fundamental problem: the barrier to building Agents.

Building a true AI Agent currently requires either Dify/n8n-type workflow platforms (which don't support autonomous decision-making) or LangChain/CrewAI-type code frameworks (which require Python programming skills). Neither approach lets non-technical users build Agents independently.

A lawyer won't use LangChain. An accountant can't configure a ReAct Agent. A marketing manager doesn't write Python.

SoloEngine fills precisely this gap.

SoloEngine is a low-code Agentic AI development platform. Users open a browser, drag Agents onto a canvas, connect collaboration relationships, configure the tools they need, and hit run. The backend automatically compiles the visual design into an executable Agentic AI system—one that plans tasks, executes operations, and delivers real-time feedback, while users only need to review and confirm.

No lines of code. No if/else logic to configure.

SoloEngine uses genuine Agentic AI architecture—each Agent runs a "think → act → observe → repeat" loop, making real-time decisions based on current conditions rather than following preset paths. Hit an unexpected obstacle, and the Agent finds its own detour. Spot a better approach, and it switches routes on its own.

Here's how SoloEngine stacks up against the mainstream options:

	Dify/n8n	LangChain/CrewAI	SoloEngine
True Agentic AI support	✗ Preset-path workflows only	✓ ReAct / multi-Agent	✓ ReAct / multi-Agent
Programming required	No	✗ Must know Python	No
Visual orchestration	Partial	✗ None	✓ Full canvas experience
Can domain experts build independently	Yes (but no true autonomous decision-making)	✗	✓
Multi-Agent collaboration	✗	✓	✓

Progressive disclosure—tools, Skills, and MCP protocols load on demand, so Agents only invoke the tools they actually need, cutting token consumption by over 85% in complex tasks. Unified adaptation layer—covering OpenAI, Anthropic, Ollama, MIMO, DeepSeek, Tongyi Qianwen, Zhipu, and all other major models. One-click packaging—assembled Agent teams can be packaged into complete products.

MiMo's 99% price cut drives Agent operating costs toward zero. SoloEngine drives Agent development barriers to zero. Stack the two together, and SoloEngine's progressive disclosure mechanism saves another 85%+ on tokens.

Take a concrete scenario: a lawyer drags a "Contract Review Agent" onto the canvas, adds a "Legal Statute Search Agent" and a "Risk Flagging Agent," connects their collaboration relationships, and hits run. Thirty minutes later, a contract review report with 37 flagged risk points is automatically generated. With MiMo's post-price-cut API, monthly costs drop from thousands of yuan to the low hundreds.

While OpenAI is still locking AgentKit into the GPT-5 ecosystem, Xiaomi has already driven the barrier to Agents down to zero with the MiMo price cut plus SoloEngine combination.

SoloEngine's positioning is crystal clear: No Workflow. No orchestration code. Just Agents that get things done.

Github:https://github.com/Sh4r1ock/SoloEngine

I just looked through the technical documentation, and this thing might actually be the kind of product that takes AI development into a whole new era. Alright, let me go ahead and deploy it so I can see what it can do.

NotAlex — Thu, 04 Jun 2026 13:49:54 +0000

Dufrence

Jun 4

60 Billion into AI: The Final Mile of Xiaomi AI Ambition

#ai #programming #python #agents

7 min read