AI-Assisted Development Workflow
Using AI tools and agents (LLM-based coding assistants) in everyday backend work, from reading unfamiliar code and writing boilerplate to drafting tests and tidying up documentation.
- Client
- Internal / Personal
- Role
- AI Tools & Agents
- Duration
- Ongoing
- Team
- 1 Developer
- Completed
- November 2026
Backend work involves plenty of repetitive tasks such as writing CRUD, test cases, and documentation. Time spent on those leaves less room for thinking about system design and tracking down hard bugs.
Using AI tools and agents as helpers in the workflow, with specific prompts and clear code context. Every result is still read and tested before it goes into the repository, so quality does not depend on AI output alone.
Project Overview
My daily backend work has a fair amount of repetition: adding endpoints, writing tests, updating API docs, and reading old code to find the source of a bug. I hand part of these tasks to AI tools and agents, then review and adjust the results myself.
What I use:
- LLM-based coding assistants to re-read unfamiliar code, explain how a function flows, and suggest small fixes.
- Agents for repetitive work, such as generating test skeletons, writing handler boilerplate, or drafting documentation from existing code.
- Translation and text clean-up for the two-language content on this site, including the CV and portfolio pages.
Principles I follow:
- Always review the output. AI can be wrong and does not know the business context. The final call is mine.
- Context has to be clear. The clearer the context (files, rules, code style), the less rework is needed.
- Not for sensitive things. I do not put credentials, customer data, or confidential production code into public AI services.
Done this way, the boring technical work finishes faster and my time can go to things that need judgement, like service design and debugging.
Everyday examples
So this does not sound like theory, here are a few real things I hand to AI and what I do afterwards:
- Understanding code left untouched for a long time. If I have to fix a part someone else wrote years ago, I ask an assistant to explain how it flows first. Only after that do I decide where the change belongs.
- Drafting tests. For endpoints whose behaviour is already clear, I ask for a set of test cases. I add the edge cases myself, such as empty input, duplicate data, or access without permission.
- Documentation from code. Field lists and sample responses are usually already in the code. The assistant helps summarise them into a documentation draft, then I adjust the wording to match what the team uses.
Limits I hold to
AI is useful for work with clear patterns, but I do not use it to decide anything that touches other people’s data. Designing access rules, choosing how sensitive data is stored, or deciding what a user is allowed to see stays with me, because those decisions need context that is not in the code.
I also do not take its output as correct by default. If a suggestion feels too neat or too confident, that is exactly the part I check first.
Interested in a similar project?
Let's discuss how I can help bring your ideas to life.