The $0 AI Stack: Free Tools to Start Building

Ellie at Willa, Editorial lead, Willa · Published August 24, 2026 · Updated August 25, 2026

You do not need a stack of paid subscriptions to start building with AI. You need a handful of free tools and a method for using them well. That was the promise of Willa's live session "The $0 AI Stack: Everything You Need to Start Building Free," led by Afua Laast, founder of the Laast Firm and lead change and enablement strategist at Airtable, where she helps companies like Disney, CNN, and FedEx adapt to new technology. Afua recently completed a fellowship on ethical AI through Columbia's Digital Futures Institute, so alongside the tools she brought a healthy dose of cost, judgment, and "should you even use AI for this?" This guide captures her full playbook.

What are the best free AI tools to start building with?

Afua's $0 stack centers on five tools: Gemini, Claude, Perplexity, NotebookLM, and Microsoft Copilot. Every one of them is free to start (Copilot is the exception: "not necessarily free, but most jobs have it"), and together they cover current information, drafting, verifiable research, learning from your own documents, and workplace productivity.

But before the tool tour, she made the framing clear: your stack is an asset or a liability. There's no neutral. Most AI tools run on the same few underlying models, trained on much of the same data, so chasing every new app on LinkedIn mostly buys you panic, not capability. As Afua put it: "This is not a Pokemon situation. You don't need to collect all the things. You just need to understand what problem you're solving for yourself."

Her fundamental rule for the whole session: AI handles the task, and humans handle the judgment.

How do I choose the right AI tool for the job?

Before opening any AI tool, Afua asks two questions:

1. What problem am I solving? Name the job that needs to be done, out loud, in one sentence.
2. What does success actually look like? Define "done" before you start.

That second question matters because, as Afua and Allison (Willa's founder, who hosted the session) discussed, many generative AI apps are modeled a lot like social media: they keep asking follow-up questions and suggesting the next thing to build, pulling you in. "You came in for a 10-minute session, and it's an hour later, two hours later, and you're still prompting," Afua warned. If you haven't defined done, you'll keep clicking, and keep spending tokens.

Her routing logic for any given task:

• Quick factual question? Go to Google, or "ask a human. Humans are still there. We're still alive. We're still thriving." Both cost less energy and money than an LLM.
• Building something (a deck, a website, an agent, a deep-dive research project)? That's when generative AI earns its place.

She also recommends "one chat, one job." LLMs re-read the entire conversation with every message. Her analogy: imagine re-reading all seven Harry Potter books every time someone asked you a Harry Potter question. Long, unfocused chats burn tokens and energy. Ask the chat to summarize, then start fresh in a new one.

What is the TRACE method for writing better prompts?

Whether you call it prompting, context engineering, or few-shot prompting, Afua says it all comes down to one question: what are you providing to the AI so it can give something useful back? A generic prompt like "write me a cover letter" forces the model to fill in blanks, which is where generic output and hallucinations come from. A more verbose, specific prompt costs you 40 words up front and saves rounds of re-asking.

Her framework, used by her clients and the Fortune 500 teams she works with, is TRACE:

• T for Task: Start from the end goal, like backwards planning. Do you want a one-pager? A research summary? Say so.
• R for Rules: Constraints, format, and the "never do this" list. Rules feel limiting, but they actually free the model up by giving it a container to work in.
• A for Anchors: Examples and reference points to match, like a past piece you're proud of, or winners in your space. AI is very good at finding patterns in examples that you might not see.
• C for Context: The background it needs to get it right. Why does this matter? What's the situation?
• E for Evaluate: How should it check its own work? Did it invent data? How confident is it in its sources?

Her worked example: you want to sound more strategic at work. Task: "Here's my draft status email. Rewrite it the way a director would write it." Rules: "Don't invent wins I didn't claim; keep it under 120 words." Anchor: how your higher-ups actually write. Context: you're up for a promotion. Evaluate: "Show me the three biggest changes and why each one reads as senior." That last step is the one most people skip, and it's where the learning happens. In Afua's words, "the rewrite becomes the receipt, and the lesson is the product."

One member shared that she uses a similar framework but had been leaving evaluation until later, and realized on the call that this was exactly what made her editing cycles so long.

What is each free AI tool actually best at?

Afua ran this section as a trivia game: clues first, reveal after. Here's the substance:

• Gemini (formerly Bard; the name means "twins" in Latin) is her pick for the most current, real-world information, and she says it has the widest context window of the group: you can feed it huge amounts of files and photos, even on the free plan, and it holds up. It's also strong for deep research. Her memorable proof point: she photographed her outfit every day for a month and asked Gemini for styling advice, and it remembered her whole closet and could answer "I'm going to the office, then a happy hour, and it's supposed to rain" better than ChatGPT could.
• Claude (from Anthropic, founded by a brother-and-sister duo who left OpenAI) is the tool that "holds your voice," and Afua's designated second-draft partner. She pushes back hard on the conventional advice to let AI write your first draft: think through your idea yourself, then feed your draft to Claude and ask it to challenge you. Better yet, have it act as your board of directors. Pick the leaders you admire and ask how each would push back. (Her cap: no more than six personas, "because even in the AI world, when you have too many voices in the kitchen, things get messy.") She also relies on Claude Projects to keep each client's work compartmentalized.
• Perplexity (founded in 2022 by a former OpenAI researcher; an "answer engine," not a search engine) is for when you need sources, not just an answer. "We love Perplexity because it gives receipts," Afua said, pointing to a recent consulting-firm report that was caught citing sources that didn't exist. Every Perplexity answer comes with clickable footnotes, you can filter toward social sources like Reddit for unfiltered customer sentiment, and answers can become shareable PDFs.
• NotebookLM (which began as a Google Labs experiment called Project Tailwind) got the deepest dive, because Afua thinks it's the tool people use least. It's a closed loop: it only knows what you upload, which makes it ideal for understanding something deeply and quickly. Load PDFs, YouTube links, or podcasts, then ask your documents questions, generate quizzes and study guides, or create an audio overview, a podcast you can actually interrupt with your own questions. Her favorite use case: job hunting. Upload the job description, your resume, and interviews with the CEO, and prep from there. If you have a Gmail account, you already have access.
• Microsoft Copilot rounds out the stack because your employer probably already pays for it. Afua's advice: "Let your job be kind of your apprenticeship." She used Claude and Gamma heavily at work before ever paying for them herself.

When should I actually start paying for AI tools?

Afua's rule: "Largely stay free until the evidence says otherwise." Upgrade only when a tool has proven repeated use for the thing you keep coming back to, and then pay for that one tool, not five. "Even if it's $20, $20 ten times becomes quite a lot of money."

Two more filters from her rapid-fire close:

• How expensive is being wrong? For a dinner recipe, skip the checks and balances. For a client deliverable with your name on it, keep a human in the loop and spend the extra time evaluating.
• Rules versus judgment. If a task is an "if this, then that" rule (a meeting ends, you send notes), automate it with AI. If it's strategy or creativity, where you want people to feel something, use AI to polish and research, but "the thinking still sits with you."

And her homework was refreshingly small: whatever you learned today, open one free tool and use it one time. Start there.

Key takeaways

• The best free AI starting stack: Gemini for current information and big uploads, Claude for second drafts and voice, Perplexity for cited research, NotebookLM for learning from your own sources, and Copilot (or whatever your job already pays for) as your apprenticeship.
• Don't collect tools: most run on the same few models. Name the problem you're solving in one sentence and define "done" before you open a chat.
• Prompt with TRACE: Task, Rules, Anchors, Context, Evaluate. The evaluate step is the one most people skip, and it's where you learn.
• Keep chats focused (one chat, one job) and start with a specific, verbose prompt to save tokens, money, and energy.
• Stay free until the evidence says otherwise; pay only for the tool you keep coming back to, and let AI handle rule-based tasks while the thinking stays with you.

If you're just starting to build these muscles, AI Skills for Non-Technical Professionals: Where to Start is the natural first step. And when you're ready to turn your $0 stack into something you've actually shipped, try building your first website with Claude Code.

Join us for the next one

This guide came from a live Willa event, where members played tool trivia with Afua, asked their questions in real time, and even got a shot at winning a one-on-one AI stack session with her. 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 start building alongside a community that's figuring this out together.

• free AI tools
• AI stack
• prompting
• NotebookLM
• Claude
• Perplexity

Join Willa, the community where women learn AI together.

All posts