Why 99% of buyers need something completely different from what the tech media is talking about.
The AI agent market is splitting into two distinct categories with fundamentally different architectures, interfaces, and use cases. Coding agents (Claude Code, Gemini CLI) are purpose-built for developers to automate code-related work. General agents (Manus, Claude Cowork, Perplexity Computer) are general-purpose platforms for non-technical users to automate everyday workflows.
This isn't a spectrum—it's a fork. The distinction matters because 99% of the market is non-technical, yet most agent discussion focuses on the 1% of developers.
Key insight: If you're not a developer, you don't need a coding agent. You need a general agent. The terminology, pricing, and capabilities are completely different.
Over the past 6 months (late 2025–early 2026), major AI companies have pivoted hard toward agentic interfaces. Companies like Anthropic, Google, and Perplexity have launched products that take action autonomously rather than just answering questions.
Why the shift? Assistants that answer questions are commoditized. Assistants that take action create defensible value. This is the next wave of AI productivity.
Definition: Terminal-based AI tools designed for developers to write, debug, and deploy code.
Examples: Claude Code (Anthropic), Gemini CLI (Google), Cursor (IDE-integrated)
Architecture: Terminal/CLI-first interface, deep integration with developer workflows (Git, package managers, testing frameworks), optimized for code generation and debugging, requires technical knowledge to use effectively.
Typical use case: A developer writes a natural language request ("Build a React component for user authentication"), and the agent generates production-ready code, tests, and documentation.
Market size: ~1% of the workforce (developers, engineers, technical founders) | Pricing: Subscription-based ($20–$200/month), flat-rate
Definition: Browser-based or web-based AI platforms designed for non-technical users to automate everyday workflows and business tasks.
Examples: Manus (general-purpose automation), Claude Cowork (Anthropic's general agent), Perplexity Computer (multi-model orchestration)
Architecture: Browser-based or web-first interface, plain English instructions (no code required), multi-step workflow automation, asynchronous execution (can run for hours or days), integration with common tools (email, spreadsheets, web apps).
Typical use case: A marketing manager says "Research our top 5 competitors, analyze their pricing, create a comparison table, and send it to the CEO"—and the agent does it autonomously.
Market size: ~99% of the workforce (operators, managers, analysts, founders, business leaders) | Pricing: Variable (subscription + usage-based credits) or flat-rate, depending on product
| Dimension | Coding Agent | General Agent |
|---|---|---|
| Interface | Terminal/CLI | Browser/Web |
| Input method | Code-aware prompts | Plain English |
| Technical knowledge required | High (Git, package managers, testing) | None (just describe what you want) |
| Output | Production code | Completed tasks, reports, automations |
| Execution model | Synchronous (immediate) | Asynchronous (hours/days) |
| Ideal user | Developer, engineer, technical founder | Manager, analyst, operator, non-technical founder |
Coding agents address a small, well-defined market: developers who want to code faster. This is a $50–100B TAM (total addressable market).
General agents address a massive, fragmented market: everyone else who wants to automate repetitive work. This is a $500B+ TAM.
Strategic implication: The venture capital and innovation are flowing toward general agents because the market is 10–100x larger.
Developers are loud, visible, and organized. They write blog posts, contribute to open source, and influence tech media. When Claude Code launches, it gets covered by every tech publication.
General agents are used by quiet, distributed, non-technical users who don't have a platform to discuss their tools. A CFO using Manus to automate financial analysis doesn't write blog posts about it.
Result: The narrative is skewed toward coding agents, even though general agents serve 99x more people.
The 1% (Developers): Highly technical, code-literate, comfortable with terminals and CLIs, optimize for speed and control, want deep integration with development workflows, willing to learn new tools.
The 99% (Everyone Else): Non-technical, prefer graphical interfaces, uncomfortable with terminals, optimize for simplicity and results, want integration with everyday tools (email, spreadsheets, Slack), want zero learning curve.
Critical insight: These are different products for different people. Trying to serve both with one tool creates a poor experience for both.
Outcome: General agent saves time and requires no coding knowledge.
The current terminology is confusing. "AI agents" encompasses everything from chatbots to autonomous systems, making it hard to discuss specific categories.
1. Coding Agents
2. General Agents
Why these terms?
Typical conflict:
A non-technical founder reads about Claude Code, thinks "This will automate my business," and tries to use it. They get frustrated because it requires coding knowledge. Meanwhile, they've never heard of general agents like Manus.
Are you a developer or engineer?
Yes → Coding agents are worth evaluating | No → Skip coding agents entirely
Do you write code regularly?
Yes → Coding agents can accelerate your workflow | No → General agents are your category
Is your workflow primarily code-related?
Yes → Coding agent | No → General agent
Do you want to learn a new interface (terminal, CLI)?
Yes → Coding agent is viable | No → General agent is essential
Bottom line:
If you're not a developer, don't evaluate coding agents. If you're a developer, don't expect general agents to replace your IDE.
Unlikely to converge: These categories serve fundamentally different users with different needs. A coding agent that tries to be user-friendly for non-developers becomes worse for developers. A general agent that tries to support code becomes confusing.
More likely: Specialization deepens. Coding agents get better at code-specific tasks (debugging, testing, deployment). General agents get better at business workflows (research, analysis, reporting, communication).
Market implication: Both categories will grow, but general agents will capture more revenue and users because the market is 100x larger.
The agent market isn't one market—it's two markets with different users, different needs, and different products.
If you're a developer:
Coding agents are worth evaluating for your workflow. They can accelerate code generation and debugging.
If you're not a developer:
Ignore coding agents. Focus on general agents. They're designed for you, and they'll save you significant time on repetitive work.
If you're evaluating AI agents for your organization:
Understand which category your team needs. Don't evaluate coding agents for non-technical teams, and don't expect general agents to replace developer tools.
The strategic opportunity:
General agents serve 99% of the market and are still in early adoption. This is where the growth is.