AI Agents & Orchestration

AI Agents & Automation

Last updated 2026-08-01

What's new

2026-08-01
  • AI can help create a virtual executive officer (a digital assistant for business tasks) using tools like Claude Code (a coding assistant) and frameworks like Seed (a planning tool) and Skill Smith (a skill-building tool).
  • To build this officer, you need to know what you want it to do, what data it can use, and how to connect it to your other software tools using MCPs (command-line tools that act as bridges).
  • The focus is on AI augmentation (using AI to improve decisions) rather than full automation (replacing all human tasks), especially if your business processes aren't clearly defined yet.
  • You can use tools like Appify (a data scraper) to gather data from platforms like Instagram and YouTube, and integrate it with your officer for tasks like competitor analysis.
2026-07-28
  • Claude Opus 5 (a new AI model) can create detailed, professional spreadsheets in Excel, like turning Nvidia's annual report into a financial model with forecasts and charts.
  • A free AI agents cheat sheet from HubSpot and Futuredia helps beginners choose and use AI agents (automated AI tools) for tasks like competitor research or organizing files.
  • To use Claude (an AI model) with Excel, install the Claude extension (a small program that adds features) to let Claude control and edit your spreadsheet.
  • Claude can also search the internet for information, like rumors about the Anthropic IPO (when a company first sells stock to the public), and add it to your spreadsheet.
2026-07-25
  • Lyft has developed a customer support AI agent (a computer program that mimics human conversation to help customers) and emphasizes rigorous offline and online evaluations to ensure its performance before and after launch.
  • Their offline evaluation involves simulated multi-turn conversations using synthetic data (fake but realistic data) and a language model (LM) judge to assess interactions, aiming to prevent using live users as test data.
  • Online evaluations include tracing tools to monitor the AI agent's actions, an online grader to evaluate its performance, and a human-in-the-loop process to analyze errors and improve the agent continuously.
  • They highlight common evaluation failures, such as creating meaningless scores, having unreliable LM judges, and lacking mechanisms to catch performance regressions (when the AI agent starts performing worse) in production.
2026-07-19
  • Thinking Machines Lab, a startup led by former OpenAI CTO Mira Murati, released a new AI model called Inkling, which is a large, open model designed to handle text, images, audio, and video.
  • Inkling is a "mixture of experts" transformer (a type of AI model) with 975 billion total parameters, but only around 41 billion activate for a typical prompt, making it faster and cheaper to run.
  • Unlike other AI models that focus on specific tasks, Inkling is a generalist, meaning it's designed to perform well across a wide range of tasks, including reasoning, coding, and following instructions.
  • Inkling is fully open, meaning anyone can download and use it for free, and it's designed to be efficient, matching the performance of other models while using fewer resources.
2026-07-16
  • Combining Fable 5 (a powerful AI model) with Hermes (an AI agent) can boost productivity and efficiency, but it's important to use Fable 5 strategically to avoid high costs.
  • A new method called "compress, judge, and execute" uses cheaper or free AI models (like ChatGPT 5.5) to gather and compress data, then uses Fable 5 to make decisions and Hermes to execute tasks.
  • A new skill has been added to help Hermes know when to use Fable 5 and when to use other models, with a cost estimate feature to help manage expenses.
  • The Agentyc Operating System (a platform for managing AI agents) now supports creating custom AI personas with specific jobs and models, like using Fable 5 as a "lead orchestrator" for big-picture tasks.
2026-07-10
  • Hermes Agent (a powerful AI tool that can act like a full-time employee) works best with the Opus model (a specific AI model that's very reliable but expensive), but ChatGPT (a popular AI chat service) and GLM 5.2 (a cheaper AI model) are also options.
  • To avoid downtime, run at least two Hermes agents simultaneously, using different AI models or accounts, so they can monitor and fix each other if one fails.
  • You can create new Hermes agents (called "profiles") either by asking an existing agent to set one up for you or by using the Hermes dashboard.
  • If you're running a serious business, consider investing in the Opus model for Hermes Agent, as it's the most reliable for completing tasks.
2026-07-07
  • New AI models like Fable 5 (a smart AI assistant) and GPT 5.6 do too much, so you need to simplify your requests to avoid overwhelming them.
  • To get the best results, tell the AI what you want to achieve, who it's for, and what "done" looks like, without giving step-by-step instructions.
  • Use a "fence" to prevent the AI from doing extra, unasked tasks by setting clear boundaries in your requests.
  • Fable 5 will soon cost extra to use, so save it for complex tasks that need high-quality results.

Key points

What it is

  • An AI agent is a program that figures out the steps to reach a goal on its own, like a construction crew building a train track from your instructions.
  • Unlike a deterministic workflow (a fixed sequence where same input always produces same output), an AI agent is more like a slot machine — pull the lever and results vary.
  • AI agents are powerful, but often overused; simple workflow automation (a reliable, step-by-step process that may use no AI at all) works better for many business tasks.
  • The real skill is knowing when an AI agent fits and when a basic automation is the right tool.

How to use it

  • Start with a clear outcome in mind and describe what you want done, instead of writing step-by-step instructions.
  • Give the agent instructions, provide access to your tools (like files on your computer or a folder of ready-made abilities), and teach the agent by showing it examples.
  • Use managed cloud agents with proper authentication rather than running tools that require you to sit there hitting "go."
  • Match intelligence to the task: use simpler, cheaper models for routine work and reserve powerful models for complex analysis.

Watch out for

  • Treat agents as ongoing tools, not finished products; AI still makes mistakes, so stay involved and review outputs regularly.
  • Avoid building everything from scratch with complex workflows or custom code; instead, think of agents as skills you assemble in minutes using plugins.
  • Never let agents use real passwords or API keys on their own; always use managed cloud agents with proper authentication.
  • Sell solutions to specific problems, not automation tools; become a long-term partner solving business pain points, not a scrapy dev selling features.

Tools named

  • Claude Code (an AI tool that figures out how to execute your goal), Apex (a tool for task automation), n8n (a drag-and-drop tool for connecting apps), HubSpot’s free AI agents cheat sheet (a guide covering seven tools, their costs, and best use cases), Co-work (a tool for AI development), Codex (a tool for AI development)

Lesson 1: What is AI Agents & Automation and why it matters

An AI agent is a program that figures out the steps to reach a goal on its own. You give it an outcome, and it decides how to get there — like telling a construction crew "build a train track from here to there" and letting them handle the details. This is different from a deterministic workflow (a fixed sequence where same input always produces same output). Think of a vending machine: put in a quarter, hit E4, get a Coke. An AI agent is more like a slot machine — pull the lever and results vary.

AI agents are powerful, but they are often overused. For many business tasks, simple workflow automation (a reliable, step-by-step process that may use no AI at all) works better and is more predictable. About half of all business automations don't need any AI. The real skill is knowing when an AI agent fits and when a basic automation is the right tool.

For AI development, this distinction matters because it guides how you build. With traditional automation, you lay every rail and switch by hand. With agentic workflows, you describe what you want and the system handles execution. Many AI agent platforms connect to your existing tools and let you create automations that run on a schedule. Successful AI development means being a problem solver first — picking the simplest reliable solution, not defaulting to agents for every task.

Sources

Lesson 2: How to use AI Agents & Automation: step-by-step

To use AI agents (programs that complete tasks for you automatically), start with a clear outcome in mind. Instead of writing step-by-step instructions, you simply describe what you want done. For example, with Claude Code, you paste your goal into the tool and it figures out how to execute it.

The process has three steps. First, give the agent instructions about what you need. Second, provide access to your tools—like files on your computer or a folder of high-quality skills (ready-made abilities the agent can use). Third, teach the agent by showing it examples. You can place system instructions (rules for how the AI should behave) in a file. If you use Co-work, that file is called Claude.md; if you use Codex, it is agents.md.

A practical example: multiple YC companies now build agents just for task automation. Tools like Apex let you say, "Here's the outcome I want," and the agent handles the rest. To match the right intelligence to the right task, you can run two different agents at once—saving money by using a simpler model for easy jobs and a smarter one for complex work.

If you want help picking what to automate, HubSpot’s free AI agents cheat sheet covers seven tools, their costs, and best use cases. Start simple, solve one problem, then expand. You do not need to be a developer—just describe what you want, give the agent access, and let it work.

Sources

Lesson 3: Best practices and pitfalls

# AI Agents & Automation: Common Pitfalls and Best Practices

Building AI agents (autonomous programs that perform tasks) sounds exciting, but beginners often stumble into predictable mistakes. The biggest pitfall is treating agents as finished products rather than ongoing tools. Many users build agents once, then expect them to work perfectly forever—they don't. AI still makes mistakes all the time, so you need to stay involved and review outputs regularly.

A critical mistake is trying to build everything from scratch with complex n8n workflows or custom code. Instead, think of agents as skills you assemble in minutes using plugins. One approach: build the information hierarchy (structured data your AI reads) first, then point any agent at it. If you switch from Claude to ChatGPT tomorrow, your hierarchy still works.

Another pitfall is security—agents cannot safely use real passwords or API keys on their own. Always use managed cloud agents with proper authentication rather than running tools that require you to sit there hitting "go." Claude's managed agents run 24/7 in the cloud and took under 30 minutes to set up.

The best practice? Match intelligence to the task. Use simpler, cheaper models for routine work and reserve powerful models for complex analysis. This saves ridiculous amounts of money once dialed in. Also, sell solutions to specific problems, not automation tools—become a long-term partner solving business pain points, not a scrapy dev selling features.

Sources
← Back to curriculum