Jianshanxing Technology

English

中文

English

中文
Sponsored by [KnowFun]: Transform your materials into personalized learning with AI.

People Who Use AI Tools Individually Are Being Left Behind by Those Who Build Systems ​

We · October 2026

This is the fourth article in the "Core Abilities for the AI Era" series. The first three covered the amplifier effect, taste and discernment, and problem definition ability. This article discusses another capability that's widening the gap: the ability to combine multiple AI tools into systems.

You've probably had this feeling:

Individual AI tools are all pretty useful — ChatGPT for chatting, Midjourney for drawing, Runway for video, TTS for voiceover. Used individually, each one boosts efficiency.

But when you actually need to make something complete — like a short video, an illustrated article, a full product feature — you realize these tools are disconnected. You have to switch back and forth between platforms, copy and paste, manually align everything. Efficiency improves, but it's still exhausting.

That's the difference between "using tools individually" and "building systems."

Most people stay at the "individual tool" stage — AI is a collection of independent features you call on when needed.

But another group of people is already doing something different: stringing multiple AI models, multiple steps, and multiple tools into a complete workflow system.

How big is the difference? Let's look at some examples.

From One Image to a Production Line: Why ComfyUI Became the Industry Standard ​

Let's start with changes in AI image generation.

Midjourney is famous, everyone knows it. But what you might not know is that among professional creators and studios, the real productivity tool is ComfyUI.

Why? Because Midjourney is "single-step generation" — you type a prompt, it produces one image. Simple, but limited control.

ComfyUI is different. It's a node-based workflow system — you can string multiple models, multiple steps, and multiple processing stages together like building blocks:

"Instead of treating generation as a one-off step, they treated workflows as long-term production assets."

— Comfy.org, Series Entertainment case study

They no longer treat generation as a one-off step — they treat workflows as long-term production assets.

What does that mean? Let's take a real example.

A game and video production company called Series Entertainment used ComfyUI to rebuild their entire production pipeline. They built a character generation system — not generating character images one by one, but building a complete production line:

  1. Base character image generation
  2. Batch output of six different expressions
  3. Style unification processing
  4. Video motion generation
  5. Final output

The entire process is a single ComfyUI workflow file. Run it once and you get the full set of results. Work that used to take a team weeks, now one person running one workflow handles it all.

Another creative studio called Groove Jones built a holiday marketing campaign for Dick's Sporting Goods. They used ComfyUI to build their own GenVFX pipeline — custom LoRAs ensuring brand style accuracy, one workflow diagram orchestrating multiple generation models, and final output of finished assets.

Moment Factory is even more extreme — they use ComfyUI for architectural-scale 3D projection mapping. Note that's architectural scale — light shows projected onto entire buildings.

"ComfyUI's node-based architecture allowed the team to design and reshape the workflow itself, not just the outputs. Rather than adapting the project to fit a tool, the tool could be adapted to fit the architecture."

— Moment Factory, ComfyUI case study

ComfyUI's node-based architecture let the team design and reshape the workflow itself, not just adjust outputs. Instead of adapting the project to fit a tool, the tool could be adapted to fit the project.

That's the value of systems thinking — you stop being a tool user and become a tool designer.

You might say: these are all professional companies, what does this have to do with me?

A lot. ComfyUI isn't some expensive enterprise software — it's open source, free, and runs on a regular computer. You can open We right now and install it in one click. The barrier isn't price — it's mindset. Can you upgrade from "I want to generate one image" to "I want to build a production line"?

From Chatbots to Multi-Agent Systems: The Next Enterprise Battleground ​

It's not just creative fields — enterprise applications are going through the same change.

The first generation of enterprise AI applications was simple — connect a large model API, build a chatbot, let employees ask questions.

But everyone soon realized: chatting alone doesn't solve real business problems.

To actually solve business problems, what you need isn't a model that can chat — it's a system that can get work done: query databases, call business systems, process documents, collaborate with humans.

And so we have Agent and multi-agent systems.

What's a multi-agent system? Simply put: instead of having one AI do everything, you have multiple AIs each with their own specialty, collaborating like a team.

In Google's 2025 Agent2Agent (A2A) protocol launch, they gave a financial operations example:

"Instead of one agent, you build three specialist agents and one manager agent... gcp-finops-agent, aws-finops-agent, azure-finops-agent — each handles one cloud platform, and the manager agent coordinates them."

— Google Cloud, The "A-Team" has arrived, 2025

Instead of one agent, you build three specialist agents and one manager agent — GCP FinOps agent handles Google Cloud, AWS FinOps agent handles AWS, Azure FinOps agent handles Microsoft Azure, and the manager agent coordinates them all.

Three specialist agents, one manager agent — together they're an "AI team."

Let's look at a more concrete example. AWS and Visa collaborated on a travel booking multi-agent system:

"A travel booking multi-agent system that plans, optimizes, and books end-to-end travel experiences on a user's behalf — bringing together discovery, personalization, and secure payments into a single, seamless workflow."

— AWS Machine Learning Blog, 2025

A multi-agent travel booking system that plans, optimizes, and books end-to-end travel experiences on behalf of users — bringing together discovery, personalization, and secure payments into a single, seamless workflow.

This isn't something a chatbot can do. It's an entire system — with planning, execution, payment, and risk control, multiple modules working together.

There's also the insurance claims example. A system built with Amazon Nova + Snowflake + LangGraph, with a pipeline that includes:

  • Document and ID extraction
  • Multimodal image analysis of accident photos
  • Policy rule verification
  • Natural language generation of claim results

The entire process is fully automatic, with humans only needed to handle exceptions.

You see — enterprise-grade AI applications stopped being "connect an API and make a chat box" a long time ago. They're building systems — stringing large models, business data, third-party services, and human review into a complete pipeline.

Why Are People Who Build Systems Ten Times More Efficient? ​

Talking about this might still feel a bit abstract. Let's do the math.

Suppose you're a content creator and you want to make a 1-minute AI short video:

Option A: Manual tool chaining

  1. Write script with an LLM → 5 minutes
  2. Copy to Midjourney to generate images → 10 minutes (selecting + adjusting prompts)
  3. Download images, upload to TTS tool for voiceover → 5 minutes
  4. Upload to video generation tool for motion effects → 15 minutes
  5. Download all assets, import into editing software to assemble → 15 minutes
  6. Add subtitles, adjust pacing, export → 10 minutes Total: ~60 minutes

And that's if you're熟练. If you're not, with反复 adjustments, it could take two or three hours.

Option B: Build an automated workflow

  1. Input topic and key information → 1 minute
  2. Workflow automatically completes: write script → generate storyboard → generate images → voiceover → generate video → assemble → add subtitles
  3. Human review and fine-tuning → 10 minutes Total: ~15 minutes

That's a 4x efficiency difference. And that's just for one video — what if you make 10 a day? What if you're making something more complex?

And systems have another huge advantage: they're reusable.

You spend time building a workflow, then you can use it over and over, continuously optimizing it. Every time you use it, efficiency improves. Manual chaining, by contrast, starts from zero every time.

That's the compound effect of systems thinking. Building the system takes time upfront, but once it's built, the output gains are orders of magnitude.

But Most People Are Still Stuck at the "Individual Tool" Stage ​

If systems thinking is so important, why are most people still at the individual tool stage?

Because there are several barriers:

Barrier One: Not Knowing It's Even Possible ​

Many people's mindset about AI is still stuck at "AI is a collection of independent features" — chatting is chatting, drawing is drawing, voiceover is voiceover. It never occurs to them these things can be connected.

It's like when smartphones first came out — most people just thought of them as "phones that can go online" — making calls, sending texts, browsing the web. But a few people saw the potential of the app ecosystem and built all kinds of applications you'd never have dared imagine before.

AI is the same. Most people are still at the "using features" stage, while a few are already "building systems."

Barrier Two: Thinking "Building Systems" Is Technical Work, Not for You ​

Many people hear words like "workflow," "system," "architecture" and assume it's for programmers, not for them.

That's not true.

ComfyUI workflows are node-based — drag and connect, no coding required. Automation tools like n8n are also visual — select trigger conditions, select execution steps, drag to chain them together. RAG tools like Dify and MaxKB are also configuration-based — select models, connect databases, set parameters, no coding needed.

The barrier to building systems keeps dropping. The question now isn't whether you can build one — it's whether you've realized you should.

Barrier Three: Thinking It's Too Much Trouble, "Good Enough Is Fine" ​

This is the most common mindset.

"I can do it manually anyway, why bother building a system?"

The problem is — you can do it manually, but you can't do much, can't do it fast, can't do it well. You manually make three videos a day; someone who built a system makes thirty a day. And the quality might be higher, because every step of the system is standardized.

Short-term, building systems takes time. Long-term, building systems saves the most time.

Self-Assessment: What Level Is Your Systems Thinking At? ​

Ask yourself a few questions:

  1. When you do something that requires multiple steps, do you manually chain it each time, or do you have a fixed process template?
  2. Are the AI tools you use connected to each other, or are they independent? (Can data flow automatically, or do you have to manually copy and paste?)
  3. When faced with a complex problem, is your first reaction "find an AI tool to solve it" or "design a process to solve it"?
  4. Have you ever "automated a repetitive process"? (Even very simple automation counts)
  5. Can you clearly describe the complete workflow of your daily work? (From start to finish, what each step is, what tool you use, what it produces)

Four Levels ​

LevelCharacteristicsTypical表现
Tool UserUses AI tools one by one, manually chains themCopying and pasting between different platforms
Process-OrientedHas fixed steps and order, but still executes manuallyHas SOPs, but every step needs human operation
AutomationHas automated every step that can be automatedWorkflow runs, humans only make key decisions
Systems BuilderCan design and optimize entire systems, continuously iterateWorkflows are assets, constantly refined and upgraded

Most people are somewhere between "Tool User" and "Process-Oriented." Those who reach "Automation" are already far more efficient than their peers.

How to Upgrade from "Using Tools" to "Building Systems" ​

The good news is that systems thinking isn't innate, and it doesn't necessarily require coding. Anyone can start practicing from simple places.

Three specific practice methods:

1. Draw Flowcharts: "Draw Out" What You Do ​

Next time you do something moderately complex (like making a video, writing an article, handling a work task), after you finish, spend ten minutes drawing out the process:

  • What's the first step, what tool do you use, what does it produce
  • What's the second step, what tool do you use, what does it produce
  • Which steps are serial (must wait for the previous one to finish)
  • Which steps are parallel (can be done simultaneously)

Draw it out and you'll discover — many steps are actually repetitive, can be merged, can be automated.

The process of drawing flowcharts is the process of turning tacit knowledge into explicit knowledge. Before you worked by feel; after drawing it out, you can clearly see the structure of the entire process, and then figure out how to optimize it.

2. Start with "Single-Point Automation": Automate One Step First ​

Don't try to build a complete big system right away — that'll intimidate you.

Start with the smallest possible step. For example:

  • You often copy AI-written text into a document → can you have AI output directly to the document?
  • You often manually rename images → can you write a simple script or use an automation tool to auto-name them?
  • You often download AI-generated content and upload it to another platform → can you connect them directly?

Automate one step first, experience the thrill of efficiency gains, and you'll naturally want to automate a second, then a third.

Systems aren't built in a day. They're assembled step by step.

3. Try Tools Built with "Workflow Thinking" ​

The most direct way to practice is to use tools that are themselves built on workflow thinking.

If you do image/video creation, try ComfyUI — no coding, node-based dragging, and you'll immediately understand what a "workflow" means.

If you do business process automation, try n8n — chaining different services together, trigger conditions → execution steps → output results, very intuitive.

If you do knowledge Q&A and RAG, try MaxKB or Dify — from model selection to knowledge base integration to workflow orchestration, everything is visually configured.

The process of using these tools is the process of exercising systems thinking. You'll start thinking from "what do I want to do" to "how do I design a流程 to do this."

And all these tools can be deployed and run locally — that's another topic, which we'll discuss专门 later.

Final Thought: Tools Are Point-Based, Systems Are Multi-Dimensional ​

Returning to our core proposition: AI is an amplifier.

Individual AI tools are individual amplification points — they can boost efficiency at specific steps.

But the real order-of-magnitude improvement comes from stringing these amplification points into an amplification chain — from input to output, every step is amplified by AI, and steps flow automatically between them without manual handoff.

Tools are point-based; systems are multi-dimensional. People who only use individual tools get additive efficiency gains. People who build systems get multiplicative efficiency gains.

And there's an interesting positive cycle here:

  • The stronger your systems thinking → the better your systems → the higher your efficiency → the more time you have to optimize systems → systems get even better
  • Conversely, if you only use individual tools → you get swamped by琐事 → no time to think about optimization → stay stuck in place

That's why the gap keeps growing. It's not because tools are unfair — it's because of differences in mindset. Some people are sharpening axes; others are building production lines.

Next time you use AI to do something, try asking yourself one more question: can this become a production line?

References: