Update 11/27/25 below. This article introduces the Hopscotch Method of Development, a practical project‑management approach for nonprofits and small teams who need to keep technology costs low while still building modern, AI‑powered products.
For our first Dusoma LinkedIn newsletter, I wanted to share how we built the Marla.AI MVP, launched three websites, and set up our entire digital architecture for under $5,000. Five years ago, this would have cost five to ten times more. Today, generative AI changes the economics of technical projects. It gives a single junior developer superpowers, and it gives non‑technical leaders the ability to write clear specifications, plan features, and manage development cycles.
This approach allows you to reserve your expensive senior specialists for short, efficient sprints while letting AI handle the bulk of the early planning, technical translation, documentation, and stuck‑point discovery. This is the transparency we believe in, and we hope it helps other organizations save money.
Team Inventory (Hypothetical Example)
- Junior developer working 20+ hours per week. Computer‑science student. Approximate cost: 20 dollars per hour.
- Organizational lead. Non‑developer with 25 years of technology-related experience. Uses GenAI tools for everything related to the org.
- CTO. More than 20 years of engineering experience.
- Senior developer A (overseas, met through Upwork). Approximate cost: 40 dollars per hour. Highly efficient, already working significantly faster with the aid of modern AI tools.
This blended team can operate at under 500 dollars per week because each person works only at the point where their contribution has maximum leverage.
How Dusoma Hopscotch Works
The Hopscotch Method is built around a sequence of hand‑offs (“hops”) designed to avoid wasted time at every level. Each person touches the work only when it matches their highest value.
Step 1: Organizational Lead Uses AI to Define the Feature
Using tools like ChatGPT, Replit, Lovable, and Claude, the organizational lead turns a vague idea such as “export your chat” into:
- a clear feature description
- a technical specification
- a lightweight project plan
- a list of required skills and libraries
This eliminates unclear requirements. It removes the expensive back‑and‑forth that usually happens when non‑technical leaders cannot articulate what they need.
Step 2: Junior Developer Attempts the First Build
The junior developer reviews the specification and decides whether they have the skills to attempt the implementation. They begin working with AI tools such as ChatGPT, Cursor, and Replit.
AI generally answers about half of what the junior developer needs. When the developer hits the edge of what the AI knows, they enter what we call “ditch mode” — spinning wheels on a very narrow problem.
The junior distills the difficulty into two or three precise questions. Not “how do I build this?” or “Can you take over?” but “I am stuck on this specific function or integration point.”
Step 3: Senior Developer Provides Precision Unblocking
Senior Developer A steps in for only two hours. Their job is to get the junior unstuck, not to take over the feature.
They may:
- schedule a meeting to discuss
- rewrite a prompt in a more technical frame
- show a missing pattern or architectural step
- translate the problem into a version AI can solve more reliably
If this unlocks the problem, the junior resumes development immediately.
Step 4: CTO Handles the Edge Cases
If the senior developer cannot unblock the problem, it hops to the CTO. This is the human wisdom step — the place where context, judgment, and long experience become irreplaceable.
The CTO typically invests two hours or less, helping frame the challenge in a way AI understands or adjusting the architecture to avoid a dead end.
Step 5: Junior Developer Completes the Feature
Within twenty‑four hours, the junior developer is back on track. The cycle repeats, with the junior gaining skills rapidly and the more expensive humans contributing only when their input has maximum leverage.
Why This Works
The Hopscotch Method minimizes senior hours, eliminates vague requirements, and trains juniors far faster than traditional apprenticeship models. It keeps weekly out‑of‑pocket costs under 500 dollars while still producing reliable, modern software.
AI is not replacing humans. It is simply reducing the number of hours required from each human, giving organizations the ability to build more with less while improving quality, transparency, and team empowerment. It also helps us work with the AI industry to identify the moments that AI fails or gets it wrong.
This method is how Dusoma built an entire digital ecosystem for less than the cost of a single month of a traditional agency contract. It is replicable, teachable, and ideal for nonprofits and early‑stage teams.
UPDATED!
This is a perfect example of the hopscotch method and how ChatGPT can help business leaders articulate their technical needs, communicate with tech people, and how a junior developer can get support and articulate when they’re stuck. This video also touches on making an AI software app with Open.AI’s API, or really using a layer on top of vanilla ChatGPT5.
https://www.youtube.com/watch?v=Y8QmfoYnqkA
This video is a behind-the-scenes look at how I use ChatGPT not just to build AI apps, but to think through strategy as a non-coder founder working on humanitarian tech. I walk through the evolution from my early GPT apps (like “What’s Wrong?” and “Drug Patents Kill Millions in Africa”) to building Marla, our own GPT-based app builder where we can finally see user conversations. That visibility matters, especially when we’re working with refugees and need to understand whether tools are actually helping.
Then I share a live example: a conversation with ChatGPT about whether we really needed to build our own software, or if we could just “skim” what people do in other tools like ChatGPT or Pi.ai. At first, the model scolds the idea as screen-scraping. But once I reframe the problem — refugees needing a portable, private “life record” instead of their data disappearing inside someone else’s AI — the answer shifts completely. We end up designing the idea of a “thin, durable capture layer” that sits alongside powerful AI systems: OpenAI does the heavy cognitive work, while our system keeps structured, user-owned records (for novels, for case management, for refugee EHR-style histories). It’s a live demo of my “hopscotch” method: using AI to climb quickly from “I don’t get it” to real technical strategy that reduces development costs by 90%.