Building Agents Like It’s 1999

The web in 1999 was weird, wild, and wonderful. Dial-up tones, Comic Sans headers, table-based layouts, and dancing baby GIFs. Everyone wanted to be “on the web” — but what that meant could be anything from a Geocities homepage to Amazon’s first clunky checkout.

By 2005, broadband and laptops made the internet feel more permanent. Flash intros were everywhere. Facebook and a fledgling MySpace gave “Web 2.0” its social gloss.

By 2007, the game changed again. The iPhone put the web in your pocket. The App Store redefined distribution. Amazon quietly launched its cloud services. And a little Australian company named Atlassian helped make SaaS mainstream.

The web wasn’t one thing. It was a patchwork of tools, frameworks, and philosophies — but at the time we just called it “digital.”

Entering the Industry

I got my start in digital around 2010 at Profero, one of the few fully digital agencies on the block. “Digital” was the buzzword of the agency world.

There were stories from colleagues about how their CEO used to make them wear sunglasses in client pitches — it was that kind of a time.

I was lucky enough to ride the wave to New York, working at Momentum, Huge, BBDO, and Droga5 as digital matured into “experience.” But through it all, “digital” remained a monolith. The word sounded exciting, but it hid the messy reality: nothing was created equal. A Flash microsite, a CMS-driven brand hub, and a SaaS product were all “digital” — but they couldn’t have been more different.

AI: Same Same, But Different

Fast forward to 2025, and “AI” feels exactly the same.

Everyone says they’re “using AI.” Investors, vendors, scammy funnel bros and business owners drop the term as if it’s a single thing. But prompting a chatbot is not the same as embedding AI in your CRM. Automating a workflow with AI enrichment is not the same as orchestrating a multi-agent system.

Right now the entire industry is building agents like it’s 1999 — bolting together raw parts, calling everything “AI,” and pretending it’s all the same.

Before and After React

The best way to understand where we are is to look at what happened next in the web’s story.

In the early 2000s, web development was duct-taped together. HTML, CSS, and vanilla JavaScript were powerful, but every team hacked things differently. Then came the atomic frameworks and CMS platforms — each promising order, often delivering confusion.

It wasn’t until Facebook released React in 2013 that a true philosophy emerged: build apps from reusable components, update only what changes, and make UI predictable. React wasn’t just a new tool. It gave developers a shared language. Today, it dominates front-end development and powers powerful tools like Framer.

As Walden Yan of Cognition put it:

“HTML was introduced in 1993. In 2013, Facebook released React to the world.

It is now 2025 and React (and its descendants) dominate the way developers build sites and apps. Why? Because React is not just a scaffold for writing code. It is a philosophy… In the age of LLMs and building AI Agents, it feels like we’re still playing with raw HTML & CSS and figuring out how to fit these together to make a good experience.”

That’s exactly where AI is today. We’re in the messy, pre-React stage — raw prompting, fragmented tools, no dominant pattern.

The AI Tool Fit Map

At Immensity of the Sea, we think the first step out of this fog isn’t waiting for “AI’s React moment.” It’s naming the categories clearly, the way web development eventually did (hand-code, frameworks, no-code, apps).

We have developed our AI Tool Fit Map, designed to do just that. It’s made up of seven distinct categories:

1. Structured, Process-First Automation

   Predictable, repeatable workflows where compliance and accountability matter.

   Examples: Relay.app, Tines, Process Street.
   
   

2. Flexible No-Code Workflow Builders

   Drag-and-drop automation for speed and breadth of integrations.

   Examples: Make, Zapier, Pipedream.
   
   

3. AI-Augmented Operational Platforms

   Core apps (CRM, docs, PM tools) with AI baked inside.

   Examples: Attio, Notion, Linear, ClickUp.
   
   

4. No-Code App Builders (User-Facing)

   Platforms that blend data + interface + workflow + AI to create bespoke operating systems.

   Example: Glide (Operating Systems, Automation Layer, CRMs, portals, dashboards).
   
   

5. Open Canvas Automation Platforms

   Developer-friendly orchestration, self-hosted if needed, deep flexibility.

   Examples: n8n, Activepieces, Windmill.dev.
   
   

6. AI-First Orchestration Tools
   AI drives the sequence of tasks, often with multi-agent collaboration.

   Examples: Gumloop, Lindy, Relevance AI.
   
   

7. Fully Agentic, Autonomous Platforms

   Agents plan, execute, and adapt strategies with minimal oversight. High risk, high speed.

   Examples: CrewAI, Autogen, OpenAI Operator.
   
   

Why This Matters for Business Owners

For founders, business owners, and ops leads, the risk isn’t that AI won’t work. The risk is choosing the wrong category of tool.

• If you need consistency and compliance, look at process-first automation.

• If you need speed, use no-code workflow builders.

• If you need depth and control, open canvas platforms fit best.

• If you need a user-facing interface, Glide is the bridge.

• If you want AI inside your core tools, choose augmented platforms.

• And if you want to push the frontier, experiment with orchestration or fully agentic platforms — but accept the risks.
  
  

Without this mental map, “AI” stays a monolith. That’s when businesses over-buy, under-use, or bolt tools together in ways that create more chaos than clarity.

Closing Thought

Back in 1999, outside of the industry nobody knew the difference between a Flash intro and an Amazon checkout. It was all just “digital.”

Today we risk making the same mistake with AI.

Until the React moment comes — if it ever does — leaders need a map of tool categories to make smart, contextual choices. Because AI is not a monolith. It’s a toolkit. And knowing which drawer to open is the difference between leverage and overwhelm.

Next Steps

I plan to expand out the AI Tool Fit Map into a deeper playbook. If you are interested, sign up for updates so you can be informed of when it drops.

If you have an immediate need to work out what tools would work best for your business, book a session and see how we can help.