Blending AI Tools to Make an Automated Business

Part of this challenge needs me to test a lot of different AI approaches to automating business. I’m experimenting, so you don’t have to.

Here’s where I’m at – it’s going to take not one AI tool, but a blend of AI tools.

AI Tooling I’ve Tried So far

I’m about halfway through my first challenge experiment (AI Directory Maker) and I have most of the first 2 agents built.

Agent 1: Existing Tools

For the first agent (AI Directory Prospector), I set myself to build this with mostly existing tools and API’s:

• Using API’s from RapidAPI like this one which returns Google Keyword data
• Using Workflow Automation tools like n8n/Gumloop
• Using AI Agent builders (still workflow tools really) like Relevance AI
• OpenAI’s API
• My own library for domain prospecting

… I had mixed results using existing workflow tools, but with all 3 tools I got it almost there. 

AI Directory Prospector: Step	AI Directory Prospector: Tool(s)	AI Directory Prospector: Status
Start: Capture initial input: user describes niche & picks best target idea.	Relevance AI, OpenAI API	Working
Explore Trends: Analyze current trends for the chosen niche or idea list.	RapidAPI Google KW API, OpenAI API, Architect API	Working
Niche Research: Gather deeper niche data and validate potential opportunities (nichereport.io)	–	Deferred
Search Engine Research: Use GPT or external APIs to explore search volume, competition, and related keywords.	RapidAPI Google KW API, OpenAI API, Architect API	Working
Make Decision: Decide whether to proceed or abort based on research data.	Relevance AI, OpenAI API	Working
Domain Prospecting: Check domain availability or auctions for relevant domain names.	My Domain Prospector Tool, Architect API, Relevance AI, OpenAI API	Working
Final Output: Provide the final result with chosen domain, idea, keyword data.	Relevance AI, OpenAI API	Working
Fail Output: Handles any errors or dead ends by reporting failure.	–	Deferred

I will continue to test tools along these lines, so if you have any AI workflow tool I must try, please do let me know in the comments.

Agent 2: Hard-Coded

Because of the stumbling blocks of this first agent I decided to experiment with hard-coding the automations for the second agent (AI Directory Builder):

• Tried different API’s, found one I can register domains with
• Retrieve initial data via hard-coded crawl, API, & AI calls
• Mix image generation AI models to create logo
• Use AI calls to establish copy, choose a color palette
• Generate & deploy final site via hard-code and a little AI vi API
• Built browser agents which can setup hosting and connect external services

This has so far been more reliable than using existing tools, but it has its issues too.

AI Directory Builder: Step	AI Directory Builder: Tool(s)	AI Directory Builder: Status
Retrieve Starting Data: Gather inputs from the Prospector Agent (domain choice, niche info, etc.).	Architect API, Search API, OpenAI API	Working – needs human sign-off
Register Domain: Register the chosen domain via DNSimple & point to hosting.	Architect API, DNSimple API	Working
Set Up Hosting, SSL, and Email: Provision hosting, configure SSL certificates, and create email accounts.	Browser Agent (expected)	WIP
Branding: Generate or apply branding assets (logo, color palette, style guidelines).	Architect API, Gemini Flash 2.0 API, OpenAI API, Scraper	Working – needs human sign-off
Build Files: Create directory website files (build from templates, do SEO groundwork).	Architect API, RapidAPI Google KW API, OpenAI API	95% Working
Deploy to Host: Deploy the built directory site to the hosting environment.	Architect API, SFTP (expected)	WIP
Connect Externals (Optional): Connect external services like Google Analytics, Google Search Console, Stripe, or set up GAlerts.	Browser Agent (expected)	Todo
Final Output: Summarize success with a live site and any relevant access details.	Architect API, OpenAI API (expected)	Todo
Fail Output: Handles any errors or dead ends by reporting failure.	Architect API, OpenAI API (expected)	Todo

Some days I’ve struggled against the idea that it’s all just too early, that the tools aren’t ready. But I will persevere.

As I look back at these first agents I can see bugs in the system, but also the slow forming of potential. This could work, eventually. For now I think our best bet is to use a blend of tools.

Using a Blend of AI Tools to Make an AI Directory Business

I’d love there to be one tool which captures every little aspect of running a business. I know there are several companies who are seeking to make ‘AI Co-founders’ and such agents which may get us there.

Right now, though, the tooling just isn’t ready.

We’re going to have to use a blend of tools to achieve this challenge – and I’m hopeful that through hammering out this diverse swarm we will uncover some really useful approaches.

To make a fully automated, (legit, value-adding), business in 2025 I’m going to blend:

• AI Browser Agents
• Straight LLM calls via API
• AI Automation Workflow tools
• Empower LLM tools via MCP
• Custom MCP servers for LLMs and for me
• Vibe coded aspects (templates, mini tools)
• Human writing/scripting/setup

I am less confident today that we can reach a 100% Automated savvy self-governing business than when I started, but I can see that we can certainly make a good attempt.

Structuring AI Tool Blends

If we’re going to get all of the above tools to play nicely together, we’re going to need structure.

Right now the AI storm is messy and unstructured. It’s too early for solid organisation to have spontaneously manifested. 

I think by the end of 2025 we’ll have a much more concrete foundation to build upon; but for now, we must make our own structure.

Structured AI Tooling:

• A new workflow schema (FlowSpec)
• Workflows running in tools like Gumloop
• ‘Holding’ LLM agents as overseers
• Our own API & ring-fenced tooling
• Centralised business docs (an ultimate `biz.json`):
  • Business aims/targets
  • Operating parameters
  • Ethical position
  • Designed Biases
  • Philosophy / World View
  • Creators preferences
• Reliable reporting:
  • Formalised human input
  • Feed of activity
  • Summarised progress updates
  • Unit testing & hallucination squashing
• Defined auto retry caps, and other limits/thresholds
• Dependency maps

Through all of this our aim is to pair the right tool with the right task. We want to minimise/remove human input, but to begin with I think we’ll need to adopt the ‘human check’ system:

By putting human input into the structure early on and accepting it, we can then record the decisions we make to start to build a replacement model.

Benefits of Blending AI tools

Apart from being essential right now, we also get some side benefits from blending tools this way:

1. Forces us to focus on structure
2. Forces us to evaluate tools often
3. Opportunity to build IP (connective tissue)
4. Diversity is resilience

Through building a structure which prioritises interchangeability of each node we build-in resilience, and begin to tame the AI storm toward our goals.

At least that’s the hope 😅.

At minimum we build up a warchest of prompts, structural logic, MCP recipes, and domain knowledge.

Risks: How long Will this Approach Last?

Blending AI tools is the right approach today, in March 2025. But how long will that be the case? 

I envisage a much more self-organising future for AI, where tooling becomes abstracted and AI agents themselves retrieve and execute extension tools autonomously.

Perhaps through this challenge we’ll even contribute towards that goal.

For now, I see the risks of blending AI as:

• Tool changes: AI tool providers are moving fast and breaking things. We can’t always be sure of reliable output
• Loss of service: AI tool companies are vulnerable to failure and acquisition, some parts of our system may need swapping out
• Redundancy: The AI storm is intense, some aspects will become redundant
• Costs: Some of these tools are costly, and experimenting with them isn’t always cheap

Blending AI Tools: In conclusion

So the challenge as I see it, is to avoid fixedness, but to still build structure. To see tooling as portable blocks and prepare systems to counter the chaos wherever possible.

In time the market will evolve, protocols, schema’s, and smarter agents will emerge. The trick is to be ready to leverage them.

My take on AI ecosystem:

In time the market will evolve, protocols, schema’s, and smarter agents will emerge. The trick is to be ready to leverage them.

— Woody Hayday (@woodyhayday) March 19, 2025

What’s your take? Do you see this approach working for my automation challenge? What am I missing? Let me know in the comments.