How to Build Your First AI Agent (No Code Needed)

Ellie at Willa, Editorial lead, Willa · Published August 4, 2026 · Updated August 25, 2026

If the phrase "AI agent" makes you feel like everyone else got a memo you missed, here's a stat worth holding onto. At Willa's live workshop "Build Your First AI Agent: From Template to Autonomous Workflow," Diana Olympia shared a graphic showing that less than 1 in 5 people have ever used a free AI chatbot, less than 1% pay for AI, and only a tiny fraction of that 1% has ever used agents or low-code AI tools. Learning even the basics puts you ahead of most people.

Diana spent most of her career in media and tech (Hulu, Viacom, A&E Networks, Foursquare), managing enterprise accounts for Fortune 500 clients like Meta, Amex, and Walmart. Today she helps solopreneurs, small businesses, and mid-size teams figure out where AI and automation actually matter for their business. In this session, she walked Willa members through her framework for automations versus agents, then built a working agent live, with no code, by talking to her computer. Here's the full playbook.

What is the difference between an automation and an AI agent?

Diana's framework comes down to two anchors (how much judgment the AI is exercising, and how much autonomy it has to work independently) and four levels:

1. Standard automation: no judgment, no autonomy. A pure if-this-then-that statement: a trigger (a meeting ends, an email arrives) kicks off a fixed action. Diana's example: when a meeting ends, her notetaker Granola sends the notes to Notion, where they're logged in her meeting notes database. Same input, same output, every time.
2. AI-powered automation: still a fixed path, but now the inputs are messy, so AI has to exercise some judgment to interpret them. Diana's example: when a discovery call ends, AI reads the raw transcript (an unstructured wall of text), extracts the pain points, her suggested solutions, and next steps, and drafts a personalized follow-up email. Within five minutes of a call ending, a custom draft is in her Gmail; she polishes it to sound like her and hits send.
3. AI agent: this is where the shift happens. "You give an agent a mission," Diana explained. You define the goal, but not the steps. "It's not only executing, but it's thinking, it's planning, and it's doing this autonomously... that's why it's called an agent, because it has agency to make its own decisions independently."
4. Agentic workflow (multi-agent): multiple agents, each with its own specialty and tools, collaborating like a team. Think of a chef and a sommelier: the chef decides the dish, and the sommelier picks the wine based on that decision.

One reassurance she offered up front: most of what you'll build, especially at the start, will not and should not be multi-agent. The bar for needing one is high.

How do I build my first AI agent step by step?

Diana built her agent live in Notion AI, dictating the whole thing with Whisperflow, a voice-to-text tool. No typing, no code. The agent's mission: assess whether a prospect from a sales call is a good fit, and draft the appropriate follow-up. Here's the process she demonstrated:

1. Open your agent builder. In Notion, she clicked the chat icon, opened Notion AI, and chose "create custom agent."
2. Dictate the mission, trigger included. Hers: when a new item lands in her Granola notes database and the meeting type is "prospect," assess the lead and score it a great fit, okay fit, or not a fit.
3. Tell it what to reference. She pointed the agent at her meeting notes, company research on the open web, and an ICP (ideal customer profile) page in Notion that lays out the signals of a good-fit client.
4. Define the action for each outcome. Great fit: draft a short Gmail follow-up promising a proposal in two to three business days. Not a fit: draft a polite decline, point them to her free newsletter, and create a Google Calendar reminder to check back in 90 days. Okay fit: flag it in her Slack #needs-review channel for human judgment.
5. Ask the agent to check your architecture. She ended her prompt by asking it to confirm this works as a single agent rather than a multi-agent, and to flag any apps it couldn't access. If you're ever unsure whether something should be one agent or several, her tip is simple: ask.
6. Answer its questions and connect your tools. The agent asked for links to the source database and ICP page, and confirmed its connections to Gmail, Google Calendar, and Slack.
7. Test it. She ran the agent on a dummy prospect meeting; it came back scored a great fit, reasoning spelled out, with a real follow-up draft waiting in her Gmail drafts folder.

Notably, she didn't oversell it: "Usually you will have to refine it... I don't want to make it look like magic, because there is some work involved."

When should I use an automation, an agent, or a multi-agent workflow?

Diana's rule of thumb is to match the tool to the shape of the task:

• High-volume, repetitive, structured, linear → standard automation. Client onboarding, invoice reminders, data entry: things you want predictable and consistent.
• Fixed outcome, but varied or messy inputs → AI-powered automation. Branched email sequences, content repurposing, FAQ handling.
• Open-ended, dynamic, involving research or analysis → single agent. Diana's own agents include a meeting-brief agent triggered by new calendar events, a daily brief, and a podcast digest that scans the shows she follows and flags relevant episodes.
• Multiple outcomes, multiple steps, genuinely complex → multi-agent. Her example: a content workflow where one agent monitors competitors and social, another writes copy, and a third decides what to post and when.

She also suggested pressure-testing with three questions: how flexible does the reasoning need to be, how much research or context is required, and how much precision is at stake? Anything touching contracts or revenue demands heavy testing and validation. Another useful lens: what's the risk if something goes wrong? A morning brief only you read is a much safer first build than anything client-facing.

What tools do I need to build an AI agent without code?

Diana deliberately avoids being prescriptive about tools, because functionally many of them do the same thing; the differences are mostly user experience. From the session:

• For automations: Zapier, Make.com, Gumloop, or Relay. Diana runs about a dozen simple three-to-four-step automations in Zapier and stays on the free tier.
• For agents: Notion is her recommendation, and it's where she built the live demo. You'll need the paid plan (the lowest tier is $20/month) to use agents, a price she called "a bargain" for the value. Zapier also builds agents now, in a separate section from its classic Zaps.
• For prompting: Whisperflow, for dictating prompts by voice. Don't worry about phrasing things perfectly; AI is good at figuring out what you're getting at, and you can always revise.

Her larger philosophy is anti-tool-chasing: she made a conscious decision to centralize on three platforms (Notion, Google, and Slack) and stop adding to her stack. "A new feature that shows up on a tool that you've never heard of, it will probably come to the tool that you know and love today," she said. Or, more bluntly: "The best system is the one that you use."

When one member asked about costs, Diana noted that Notion is expected to move to a credit system for agents, one more reason to keep your first builds simple.

How should I actually get started this week?

Diana closed with a loose beginner's plan, and it starts smaller than you'd think:

1. Identify one task. Not the linchpin task that drives revenue, but "one annoying, pesky task" you've punted three times because you dread it. Those are perfect first candidates: low stakes and narrow. Automating your meeting notes is a great example.
2. Build it, test it, refine it. Then observe. Sometimes you can't say "this saved me 10 hours," Diana noted, but you can say it removed the dread of getting started, and that still counts.
3. Add a level. Maybe a week later, upgrade the automation to interpret data or customize an output. A week after that, build your first agent, or start today if you're ready.

She was equally honest about what comes after: agents are easy to build but not always easy to maintain. Models can experience "context drift," gradually generalizing beyond your original parameters until quality dips, at which point you recalibrate. "It's just like tires," she said. "You need to check for pressure... you get the tires rotated." Her bottom line was blunt: "Ignore the hype, start small, keep it simple. That's a hill I will die on every single day."

And when you build your first one, share it. Diana's favorite part of these workshops is hearing from people who just shipped their first agent. "No one is an expert on this, I can guarantee you, no matter how much they say they are."

Key takeaways

• Automations follow rules; agents get a mission. The anchors that separate the four levels (standard automation, AI-powered automation, single agent, multi-agent) are judgment and autonomy.
• You can build a real agent with zero code: Diana built a lead-scoring agent live in Notion AI by dictating the trigger, the references, and the action for each outcome.
• Match the tool to the task: repetitive and structured → automation; messy inputs, fixed outcome → AI-powered automation; open-ended and research-heavy → agent. Save multi-agent for genuinely complex, multi-outcome work.
• Start with one annoying task you keep punting, build the simplest version, and notice the time saved or cognitive load lifted before adding complexity.
• Agents need maintenance: watch for context drift and recalibrate like you'd rotate your tires. Ignore the hype, start small, keep it simple.

If this sparked ideas, two natural next steps from the Willa library: see how these building blocks combine into a full AI personal assistant, and if you're watching your budget, start with our guide to a $0 AI tools stack before committing to paid plans.

Join us for the next one

This walkthrough came from a live Willa event, where members watched Diana build the agent on screen and asked their own questions in real time. Willa hosts expert-led sessions like this almost every week for women building careers with AI. See what's coming up on our events page, or become a member and build your first agent alongside a community that's figuring this out together.

• AI agents
• automation
• no-code
• Notion AI
• workflows
• beginners

Join Willa, the community where women learn AI together.

All posts