Software-shaped work
Citizen developers improve their own work with software, automation, and agents, whether or not they have a developer role. A good first project begins with a recurring task, a result you can recognize as correct, and a person who will review what the system does.
Use this guide to classify the opportunity, decide who will use the solution, and choose a supported TritonAI path.
Frame the project before choosing a tool
Work pattern
The first choice is the relationship between the software and the work you do today. Each pattern can use deterministic code, an AI model, an agent, or a combination.
Keep the job and output the same while software handles repeatable steps. Start here when you already know what correct looks like and can return to the manual process if needed.
Example: Turn a recurring spreadsheet cleanup into an inspectable pipeline that previews changes before export.
Keep the purpose of the work while giving people a more useful result. The new output might be interactive, current, searchable, or available without another meeting.
Example: Replace a recurring status deck with an accessible page that shows when its source data was refreshed.
Create a bounded capability that was not feasible as a manual job. Treat adoption and usefulness as questions to test because there is no proven process to copy.
Example: Monitor approved public sources for specific changes and route relevant findings to a person for review.
Delivery class
Durability, access, support, and testing increase as a build moves beyond its original creator. A temporary build still follows UC San Diego requirements for data, security, privacy, and appropriate use.
Answers a question or tests an interaction with approved, representative material. Keep it isolated, record what it proves, and stop or revise it when the test ends.
Handles a recurring need for a known, bounded group. It needs reliable results, an approved workspace, a review step, and a way to recover when it fails.
Serves people beyond the builder's team. Add a named owner, accessibility, privacy and security review, realistic-load testing, support, monitoring, and change management.
Serves an open or general audience. This requires a formal product and service process with ongoing ownership. It is rarely the right first citizen-development project.
Opportunity scan
Look for work you repeat, explain, reconcile, or review. The examples below can start a conversation. Each still needs the appropriate review before work begins.
Turn a repeated explanation into an accessible walkthrough, comparison, calculator, status page, or interactive explainer.
Build a reviewed pipeline that turns approved source material into a draft summary, update, brief, or set of channel-specific versions.
Reconcile exports, repeat an analysis, surface exceptions, or replace a recurring snapshot with a source-backed dashboard.
Extract approved fields, fill a template, convert files, or link a finding back to the passage a reviewer needs to inspect.
Organize submissions, identify missing information, route a request, or prepare an unsent response for a person to check.
Assemble a briefing, monitor a deadline, update a checklist, or coordinate repeatable steps across approved systems.
TritonAI ecosystem
Begin with the simplest supported option that can meet the need. A workflow may move to a different path as its audience, data, or support requirements change.
Interactive agent work
Explore a bounded task across files, browser work, and supported connections. Keep approval points visible while you learn what should become repeatable.
Repeatable workflow
Run defined steps from a schedule, message, webhook, or file event when the process and its exceptions are understood.
Application or shared capability
Add approved model access to software you operate, or package instructions and reference material so a reviewed capability can be reused.
First project
A useful starter project has a known result, limited access, and a clear fallback. Write down the answers below before opening a builder or asking an agent to act.
What do you do often enough that improving it would matter?
What does a correct output look like, and what source can you check it against?
Which people, systems, and data are in scope? Which are excluded?
Who inspects the result before it changes a record, reaches another person, or informs a decision?
Can you stop the workflow and return to the current process without losing access or data?
Who decides whether to keep, change, support, or retire the build?
Training conversation
Bring a short problem statement to a citizen-developer session: what happens today, who does it, what goes in, what comes out, how often it happens, who reviews it, what data it touches, and what happens if it fails. That is enough to choose a pattern and test a first version.
This page adapts the build framework in The AI Daily Brief episode "How to Start AI Coding If You Haven't Yet" for the TritonAI ecosystem. UC San Diego's campus data and service requirements apply to every path.
Start with a bounded task and approved data. Keep the current process available while you test the result.