Essential AI Agent Capabilities to Run a Business

Essential Skills for AI Agents and AI Agent frameworks - for my 100% AI business challenge

Building an AI-run business from the ground up: Core technical skills I need my agents to have

New here? Welcome! This is the journey of building a 100% automated AI business in 2025. You’re jumping in after we’ve already kicked things off, so you might want to catch up first.

Check out these key posts to get the full story—and don’t forget to subscribe for updates and exclusive perks:

Can a business run entirely on AI? That’s the challenge I’ve set for myself—designing a network of AI agents to handle everything from communication to decision-making. But before I start, I need a checklist of essential skills my agents must master.

Last week I laid down my AI Business Blueprint. This week I’ve been distilling my fragmented notes into a requirements scheme so I can start building:

How many senses does a robot have? – AI Senses

Humans rely on at least five senses to understand the world. AI agents, in contrast, ‘see’ data streams—webpage html, json from APIs, results from databases, or other docs. While they won’t touch, smell, or taste in 2025, their ability to process vast amounts of information at lightning speed is their true superpower.

  • Faster access to all human digital data
  • Fresher access to live data
  • Wider access to business signals
  • Very fast multi agent inference via AI agent frameworks
AI Agent framework for absorbing different data sources - essential skills for AI Agents

…in this way I hope to explore a new thing (a 100% AI business), rather than joining in the conversation about AI agents replacing human workers. Abundance mindset, always.

AI Agents Data Sight – Consuming Data

These Agents’ primary sense will be data absorption. Sometimes this’ll be documents, other times it’ll be competitors websites, customer feedback, emails, or feedback messages from me. 

Each business or project managed by AI’s might have it’s own ‘stack’ of needed input capabilities, but here’s what I want my AI agents to be able to absorb freely:

  1. Web – Search Goog, find websites, view websites, take screenshots, analyse multiple pages and discern key information from them
  2. Google Doc’s – Navigate a shared Google drive, read documents, spreadsheets and other files to inform decision making, asset generation and more 
  3. Email – Read incoming mail from humans or other bots
  4. API’s – Being able to consume API endpoints will allow me to easily open up access for each agent to read many data sources
  5. Telegram – By being able to read messages from a messenger app, agents will be able to interact with each other, and humans (may also use Slack or Matrix)
  6. MySQL – Direct ability to query a database will allow performant shared data generation and analysis between multiple agents

There are probably solutions already out there which offer most of these, but do they do it well? Are all available to one AI agent or workflow? This’ll be the first thing I look for – I’ll share what I find with you soon.

Agent Tools: Absorb Data from multiple sources (AI Agent capabilities)

Are there any I’ve missed? What would you want in your AI agent input stack? Let me know in the comments.

AI Agents Communication Skills – Inter-agent and AI<>Human

Once we have AI agents able to absorb information from our key data silo’s, (which can include instructions for them on what to do, when, and how to do it), next they need to be able to communicate outwardly.

…for the sake of this challenge, primarily in written English. A really exciting benefit of using AI in 2025 is that language barriers are falling away; but for the sake of keeping this challenge as simple as possible (😅) ​​we’ll stick to English.

Technically, I expect this to take place via:

  1. Everything logged into a MySQL database (to allow backtracking)
  2. AI agents talking to each other (multi agent, essentially via API, or maybe Slack)
  3. Emailing replies
  4. Creation/Updates of reports/docs (Google docs)
  5. Posting to online support tools/comments sections
  6. Reporting to me (business owner) via Telegram/Slack

AI Agents ‘Arms’ – Other Actions

As well as communicating with each other and with humans, these bots will absolutely need to be able to ‘act’. This is probably where the value-added will be, and I suspect, it’ll be the most likely pain point for me as a systems engineer.

  1. Code & deploy – Interact with Git repos, reading issues, writing code, submitting pull requests, reviewing each others changes, merging, deploying
  2. Automate transactions – Issue refunds, make payments (Stripe API & Pay for services via credit card 😬)
  3. Database Management – Query and modify MySQL databases efficiently
  4. Run scripts (SSH, PHP)
  5. Upload files to servers (SFTP)
  6. Send POST API requests to third party services
  7. Schedule future actions
  8. Spawn AI agents, Shutdown AI agents
  9. Wave their AI hands at me when it all falls over or explodes (Telegram message)

AI Capabilities Needed to Run a 100% AI Business

Putting all that together, I’ll eventually want my AI swarm to be able to outwardly ‘do’ the following via an AI agent framework: 

  1. Browse the web (search Goog, view web-pages)
  2. Google Doc’s (navigate google drive, read & write documents, spreadsheets, etc.)
  3. Email (read, reply, forward, and send email)
  4. Interact with API’s (consume & request, e.g. Stripe refunds, generate images, or analyse search engine data)
  5. Telegram (or Slack – talk as a business group, report to me)
  6. MySQL (direct access to a database)
  7. Log everything (probably via MySQL queries or API)
  8. Talk to other agents (via API, or Slack)
  9. Software engineering (interact with Git repo’s, github.com, and code bases)
  10. Schedule future actions (via my own, or another scheduling system)
  11. Spawn AI agents, Shutdown AI agents
  12. Run scripts (SSH, PHP)
  13. Pay for services via credit card 😬
  14. Upload files to servers (SFTP)
  15. Posting to online support tools/comments sections
  16. Wave their AI hands at me when it all falls over or explodes (Telegram message)

I’ll need the AI agents to use these actions:

  • Sensibly (not refund a customer $1000 for their $10 purchase)
  • Securely (not accidentally send customer SFTP credentials)
  • Sustainably (not create 1,000 clones in 1 minute to do the same task)
  • Understandably (we’ll need to log everything and be able to work out the kinks)

How Tho?

In the coming weeks I’m going to build such agents & agent tools, and then we’ll be ready to start our experiments. I’m not sure which platform or format will work best, but I’ll share everything I find out as I explore.

Next week I’m going to use some AI tools I’ve previously made to explore potential business models which may be our ‘lowest hanging fruit’ to achieve this challenge.

AI Agent framework - the AI Agent tools I'll need each of my agent employees to use to keep business profitable

Note, this list is somewhat personal. I like to run certain businesses where I have domain knowledge, (software/web-based). I also prefer certain tools/apps. This post represents my ‘preferred stack’. I appreciate all feedback in the comments!

My Question to You:

I’m diving into building AI agents that can run an entire business. But I know I’ll miss something. What key AI capabilities would you add? What business model would you bet on? Drop your thoughts in the comments—I’m eager to hear!

… and if you haven’t subscribed yet, let’s go already!

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *