Building the Right Tool: Prompt-to-app Workflows in Marketing Agencies

For years, the standard fix for a marketing team operational bottleneck has been “more process,” resulting in rigid, step-by-step instructions designed to force everyone through the same narrow funnel, in the hope that it would reduce churn.

The reality? We ended up with workflows full of manual workarounds, disconnected tools, and a lot of admin effort.

That era is ending, thanks to the rise of AI-assisted coding, also known as “vibe coding”. At Chemistry, we’re leaning into our spirit of experimentation and building internal solutions that greatly reduce manual effort and do precisely what we need, ensuring our tools evolve at the speed of our ideas.

What is a Prompt-to-app Workflow?

Essentially, it’s a workflow where professionals, who might not be trained developers, use Generative AI agents as their primary coding partners. Instead of writing programming-language-specific syntax, they provide natural language prompts describing the desired outcome, the logic, and the “vibe” of the interface. The AI handles most of the technical heavy lifting, writing the actual code in real-time.

As Andrej Karpathy, former Senior Director of AI at Tesla, noted, “The hottest new programming language is English.”

For agencies, this means the barrier between “I wish a tool did this” and actually owning that tool is starting to dissolve. We’re shifting from just accepting a lot of manual effort in our daily work, to moving faster towards delivery for our clients.

The Business Case: Strategic Ownership and Efficiency

Like any AI experiment worth running, we start with intention. Every tool we build has to earn its place by making the business more efficient, the work better, or the client experience stronger. And in our experience, prompt-to-app workflows do just that.

1. Increased Profitability and Operational Efficiency

A big advantage of these custom-built tools is the elimination of the friction found in generic software. When an employee has to perform several steps in a generic tool to complete a task, you’re losing out on time that can be focused on the client.

Prompt-to-app workflows allow us to create “surgical” tools, applications designed to solve a specific agency bottleneck accurately. By shaving hours off weekly tasks, we increase operational efficiency, allowing our teams to focus on strategy and creativity.

2. Time and Money Savings for Clients

When internal operations are streamlined, overhead decreases. These savings can be passed to the client or reinvested to deliver higher-quality work.

3. Intellectual Property Ownership

When you pay for software, you’re buying into another company’s roadmap. When you build an internal tool, you own the code. You’ve developed proprietary intellectual property that makes your company uniquely efficient.

Our “Secret Sauce” Development Stack

Rather than consuming internal developer resources that should be focusing on client work, non-development teams are becoming proficient with an accessible development stack that enables vibe coding.

Here are some of the tools we’re using to turn ideas into apps:

• Claude (Anthropic): Currently, our preferred LLM for complex logic and structural coding. Claude acts as our senior architect, helping us structure the app and debug complex problems through natural language conversation.
  
• Model Context Protocol (MCP): Allows AI to communicate with our existing tools like Figma to leverage design data when creating internal and client-facing interfaces.
  
• AI Agent + Visual Studio Code (VSCode): For those more knowledgeable in coding, this pairing helps with autocompleting lines of code or even writing and refactoring entire code blocks, all while greatly speeding up the process.

Working Smarter: Case Studies in Efficiency

We didn’t adopt prompt-to-app workflows because it was trendy; we adopted them because it solved real frustrations. Here are three examples of proprietary tools we developed to speed up internal processes, improve our work, and help us focus more on what matters.

1. The Figjam Transcript Importer

The Problem: Our strategy teams spend hours analyzing interview transcripts. They wanted a way to instantly turn transcript text into sorted, categorized Figjam sticky notes to speed up synthesis. There wasn’t a software solution or Figma extension for this at the time.

The AI-assisted Solution: In a short time, using Claude and VSCode, we developed a Figjam plugin. Our strategists can now paste an interview transcript and click “Import.” The plugin instantly parses the text, creates color-coded sticky notes for each topic, analyzes sentiment, clusters similar items, and imports them into the active Figjam board.

The Impact: Synthesis time is greatly reduced, and researchers can save an average of 2-3 hours per transcript. 

2. The Presentation Coach Agent

The Problem: We pride ourselves on flawless client presentations. We wanted a way to help all staff practice their delivery and receive objective, constructive feedback.

The AI-assisted Solution: We built a browser-based solution that provides real-time presentation feedback. Users invite a “participant bot” to their meeting and it listens in. The agent analyzes the audio, providing a scorecard on presentation speed (WPM) and usage of filler words (“um,” “ah,” “like”). An AI-synthesized “coach” provides instruction on how to improve based on the data.

The Impact: Our teams can “coach” themselves on their own schedule and improve the delivery of client presentations.

3. Cookie Audit

The Problem: With increasing global privacy regulations, we must ensure our clients’ sites are compliant. Existing cookie audit tools are often locked behind paywalls or services. We run audits often, and we needed something more consistent than manual audits. 

The AI-assisted Solution: We envisioned a tool that would use headless Chrome to load a site and provide a baseline cookie output. Since scripts are dynamic, the tool gives time for the code to settle before providing the output. It also detects tracking scripts that are needed to formulate complex privacy legal language. With the use of tag managers, it may not be obvious what products are tracking user behavior on any given website.

The output includes CSV or JSON data, the Chrome HAR file, which is important for legal teams, and an option to format output for Slack to drop information to teams working on the data.

The Impact: We now provide complimentary, rapid cookie audits as part of our onboarding, adding immediate value to clients at nearly zero internal cost.