Demystifying AI Tool Creation: A Practical Guide to Building Marketing Workflows

The prevailing advice to "learn to build," "ship your own tools," and "automate your busywork" often leaves aspiring builders without a clear starting point. Canva’s recent initiative, where 5,000 employees were given a full week to explore AI, highlighted this gap. The company reported that a significant portion of these employees struggled to even begin, as many AI tools assume a developer’s proficiency in file management, terminal usage, and server deployment. However, building custom marketing tools and workflows does not necessitate such technical expertise. This guide outlines a practical setup that runs entirely within a web browser via Letaido, eliminating the need for installations, file management, or continuous laptop operation.

Letaido is presented as an AI workspace accessible through a browser, capable of writing code, storing data, hosting tools, and running scheduled automations. This platform aims to empower users to create marketing tools and workflows, such as reporting dashboards, competitor monitors, keyword clustering tools, and content workflows, without relying on dedicated development teams or significant budgets. The core concept is to describe a workflow as one would explain it to a new hire, with the AI translating that description into actionable code.

The platform offers two primary environments for building: a private "Console" for internal tools and dashboards, accessible only to a team, and a public "Site" for publicly shareable pages. Several key features distinguish Letaido. It supports multiple AI models, including Claude, GPT, and Gemini, allowing users to switch between them on a per-task basis without needing separate subscriptions or API keys. The platform boasts over 35 native connectors to popular marketing tools like Slack, HubSpot, Notion, GitHub, Linear, Mailchimp, Stripe, WordPress, and Airtable, simplifying data integration.

Furthermore, Letaido includes pre-built "marketing skills," a library maintained by Ahrefs, offering agents pre-defined steps based on best practices to initiate complex tasks. Designed for team collaboration, any setup, including skills, connectors, memory, and apps, becomes available to the entire team, fostering collective progress. Free seats for team members encourage broad adoption, with role-based permissions (members, owners, admins) managing access and security.

Sharing capabilities extend beyond the team. For one-off needs like client reports, users can generate a guest link for their app or report, creating a public site with a default Letaido domain. Custom domains can also be configured, with Letaido handling security certificates. Hosting and secret management are automated, mitigating beginner risks like exposed API keys or accidental public reports. Users are advised to designate internal data for the Console and public-facing content for the Site, exercising caution before publication. For users wishing to follow the setup guide, an active Ahrefs account is required, as Letaido accesses Ahrefs project data through an existing plan. The $99/month subscription covers the Letaido platform and AI credits, but data limits are dictated by the user’s Ahrefs plan.

Setting up Letaido is presented as a streamlined process, eschewing traditional development environments like editors or Git, and eliminating the need to paste API keys. The initial step involves choosing an AI model and mode. Letaido’s subscription includes $50 in AI credits, enabling users to experiment with different models based on quality, speed, and price. The platform provides a table illustrating potential costs per run for various models like GPT-5.4-mini (3¢, ~1,600 runs/month), Claude Sonnet 4.6 (24¢, ~200 runs/month), and Claude Fable 5 (38¢, ~130 runs/month), along with their recommended use cases. Claude is favored for tasks requiring nuanced understanding and logical reasoning, while cheaper models are recommended for high-volume, repetitive tasks.

Adjacent to the model selection is a mode selector, defining the agent’s build mode, with "Auto" serving as a suitable default. Users are cautioned that credit consumption can be high when using premium models for simple tasks or when a single chat thread spans multiple unrelated objectives. Starting fresh chats for distinct tasks and utilizing cheaper models for basic operations are advised strategies.

Creating a knowledge base is crucial for providing context to the AI. Letaido offers two methods: "Memory" files (.memory.md) for essential, frequently accessed notes like recurring vocabulary, stylistic rules, or pointers to the wiki, and a "Team Wiki" (wiki/) for a more extensive workspace knowledge base, ideal for strategy documents, Ideal Customer Profiles (ICPs), competitor analysis, and product FAQs. Memory files are automatically read at the start of every chat, incurring token costs per session, while wiki pages are read only when specifically requested, with costs applied only to accessed pages. The rule of thumb is to keep Memory concise and focused on information needed in every interaction, and the Wiki comprehensive for occasional, in-depth reference.

The platform utilizes five key terms: skills, apps, reports, artifacts, and automations, each with distinct functions. A "skill" is a reusable set of steps and output formats triggered by natural language requests, with the option to clone from the Ahrefs library or create custom ones via chat. An "app" is an interactive tool built by the agent that can access databases, call connectors, and display results, suitable for monthly reports or writing workflows. A "report" is a live, read-only data view that refreshes upon access, such as sentiment overviews or lead-quality breakdowns. An "artifact" is a static, one-time output, like a historical chart or visualization. Apps, reports, and artifacts can be designated for private Console access or public site publication. "Automations" are tasks that run autonomously on a schedule, either created by describing the desired action and timing to an agent, or manually configured with limitations. The guiding principle is to package recurring know-how as skills, build interactive elements as apps, and schedule processes as automations.

Integrating external tools is facilitated through connectors. These allow the agent to interact with services like HubSpot, Apify, and others, handling endpoints, parameters, and authentication seamlessly. Connectors can read data and also perform actions, such as posting to Slack, publishing to WordPress, or updating HubSpot deals. A comprehensive list of available connectors can be found in the Letaido documentation. Adding a connector typically involves initiating a new chat and following prompts to establish access.

The process of building a workflow in Letaido involves four core steps: writing the prompt, mapping data sources, sharing with the team, and starting with a familiar workflow. Effective prompt writing is paramount, as prompts are executed repeatedly. Key elements include clear instructions, specific context, and desired output format. The best practices for prompt writing emphasize clarity, specificity, and providing examples.

Mapping data sources requires identifying all tools, apps, and sites used for work and detailing the specific data points they contain. This data can be integrated into Letaido through three primary methods: connectors, API keys (for tools not natively supported), or web scraping for public web pages.

The guide concludes with a "TLDR" summarizing the key steps: Define your objective, map your data, connect your tools, build your skill/app/report/automation, and iterate. It emphasizes that building should commence with a single, context-specific workflow rather than an extensive suite of tools.

A "Reality Check" section acknowledges the current hype surrounding AI, clarifying that these builds will not replace strategic thinking, taste, or relationships, nor will they guarantee immediate top rankings. The initial versions are expected to be rudimentary. However, the core benefit lies in automating repetitive tasks, freeing up time for more strategic and enjoyable work. The leverage comes from connecting steps into automated workflows, enabling tasks to run on command or on a schedule. The advice is to leverage existing job knowledge, start with a small build, and refine it to meet desired quality standards. Users are encouraged to share their builds upon completion.