AI has become a regular part of our workflow, and it noticeably speeds up how projects move from early planning to real development. It helps organize complex information, speeds up preparation, and takes over work that would usually slow the team down, whether it is routine tasks or larger pieces of analysis that require time but not creativity. This gives the team more space to focus on decisions that actually affect the product.
To show how this works, I chose a routing system we have built more than once for logistics products. It is a realistic example because it combines business rules, logic, interface structure, testing, and many situations that need clarity. AI does not build the system for us, but it supports several key stages and keeps the project moving at a steady pace. It helps sort information early, speeds up technical and design preparation, and reduces the number of slow manual steps that often delay work like this.
How AI Fits Into Our Development Workflow
AI works inside the process, not on the side. It helps where information is messy, where decisions depend on structure, and where preparation normally takes longer than the engineering itself. It supports research, requirements clarification, early drafts, testing scenarios, and code reviews. It does not change the core of the work, but it removes delays that used to interrupt the flow.
This structure appears in most of our custom software development projects. The steps stay similar, but the amount of help we get from AI depends on the complexity of the solutions.
What AI Changes and What It Does Not
AI changes the speed of early understanding. It helps sort notes, spot missing rules, test different logic paths, explore alternatives, and clean up information that arrives in an incomplete form. It also speeds up verification and small checks that used to take more time than they should.
AI does not replace engineering judgment. It cannot decide how a system should behave, how exceptions should work, or how interactions should feel in real use. It supports these decisions but does not define them.
A Real Example. Building a Route Optimization App
This kind of routing system appears in delivery platforms, fleet management software, and sometimes as a standalone app. It combines business rules, logic, user workflows, data handling, and reliability requirements. The goal is always the same: build something that works with real operations, not just theoretical paths. The process below shows how we approach it and where AI helps us move faster.
Stage 1. Understanding How Routing Works in the Client’s Daily Operations
We start by learning how routing actually works in the business. We ask what must happen first, what cannot change, what counts as a good route, and how drivers update plans during the day. People often describe the workflow in a simplified way, so we look at the real version early.
What AI tools we use
ChatGPT, Claude, and GitHub Copilot Chat help sort long notes into structured rules and highlight missing or conflicting details that need clarification.
Summary
We get a clear understanding of the real workflow. This becomes the base for the logic.
Stage 2. Turning the Rules Into Simple Logic
When the workflow is clear, we define how the system should react to different situations. These include standard cases like priority stops and edge cases like delays or added tasks. We keep the logic simple at this stage because complexity grows fast once exceptions appear.
What AI tools we use
ChatGPT and Claude help turn the workflow into step-by-step logic, reveal contradictions, and map out situations that need a decision before development starts.
Summary
We turn operational rules into logic the engineers can implement without guessing.
Stage 3. Preparing the Structure of the App or Module
Before writing any routing behavior, we create the structural frame. This includes screens for adding and editing stops, a place to run optimization, and simple placeholder layouts. The goal is not polish. The goal is to give the team a working environment for testing logic.
We also use AI inside Figma to generate quick layout options so the team can see early variations without spending time on full design drafts.
What AI tools we use
Copilot, Cursor IDE, ChatGPT, and Figma AI Assist help generate draft endpoints, basic data structures, temporary UI, and quick layout variations so the team can begin testing quickly.
Summary
We build the frame that will hold the routing behavior.
Stage 4. Building the Routing Behavior
This is the core of the system. The engineers write the logic that decides how routes are built. We cover incomplete data, inconsistent rules, sudden changes, and unusual situations that appear in real operations. The decisions here define how reliable the routing feels.
What AI tools we use
ChatGPT, Claude, and Copilot assist with supporting tasks like reviewing helpers, clarifying reasoning, or exploring small variations. The algorithm and logic itself are written by the team.
Summary
We turn planning into real behavior.
Stage 5. Testing Real Life Situations
Next we test how the routing behaves in everyday use. Real routing includes overlapping delivery times, last-minute changes, delays, missing addresses, and ambiguous priorities. The system must handle more than the ideal path.
What AI tools we use
ChatGPT and Claude help generate scenario variations that cover normal and unexpected situations. The team adjusts them to reflect real conditions.
Summary
We make sure the routing works outside perfect conditions.
Stage 6. Reviewing Performance and Polishing
When the routing works, we review speed, stability, and clarity. This includes performance checks, interaction adjustments, and error handling. These improvements make the routing stable enough for real operations in the app or as an independent module.
What AI tools we use
ChatGPT, Claude, and Copilot help review logs, explain difficult error chains, and highlight potential performance issues.
Summary
We finalize the routing system.
Final Result of the Process
By the time we finish all these stages, we end up with a route optimization app that behaves correctly in real conditions, not just in ideal scenarios. AI helps clear up messy inputs, check logic faster, and try different paths without slowing the team down. It does not replace the process, but it makes it move at a better pace and leaves more time for the decisions that actually shape the product.
AI Tools We Use Across the Product Team
| Area | Tool | Why We Use It |
| Working with requirements and workflows | ChatGPT | Sorts notes, reveals missing info, structures rules |
| Exploring system logic | Claude 3.5 | Strong reasoning and edge case analysis |
| Technical and market research | Perplexity | Fast research with credible references |
| Writing and refactoring code | GitHub Copilot Chat | Helps with suggestions, clarity, and small fixes |
| Creating scaffolding and prototypes | Cursor IDE | Speeds up initial setups and experiment environments |
| Understanding unclear issues | ChatGPT | Explains errors or behaviors in simple terms |
| Generating test scenarios | ChatGPT | Produces functional and edge case variations |
| Reading long logs | Claude 3.5 | Handles large text and finds patterns |
| Documentation and planning | Notion AI | Summaries, notes, task outlines |
| Early design exploration | Figma AI Assist | Generates layout variations for early review |
What This Means for Our Workflow
AI speeds up the work inside our team. It lets us build products for our clients faster, with higher quality, and at a lower cost than before. If you are starting a new product or want to improve an existing one, we’ll be glad to help your business move forward with more confidence.
FAQ
When did you first start using AI tools?
I don’t remember the exact moment because it wasn’t a big event. It started with small experiments somewhere before 2020, mostly out of curiosity. At that time, the tools were not very helpful, so we used them occasionally and didn’t rely on them for real work. Things became interesting only a couple of years later, when the tools finally became good enough to sit inside the workflow instead of slowing everything down. That was the point when they stopped feeling like toys and started becoming useful.
Have AI agents decreased the product development price at your company?
A little, but not in the way people expect. The tools don’t suddenly make a project cheap. What they do is shorten some of the slower parts, like early requirement cleanup, first drafts, or creating variations for testing. When those things take less time, the whole project moves with fewer delays, and that can reduce the total cost. The actual engineering work still requires people, so the price is mostly shaped by the complexity of the product, not the presence of AI.
Can I build my product without a development team using just no-code tools and AI design generators?
You can build a lot with no-code tools and AI design generators, including working internal tools and even early versions of a product. The limits appear when the system needs logic that goes beyond the template, when it must integrate with other services, or when reliability becomes critical. That is usually the moment companies bring in engineers, not because no-code failed, but because real operations need more control than these platforms can offer.
Will AI replace engineers someday?
I doubt it. AI already handles a lot of repetitive work, and it will probably handle even more over time, but the part where you understand the problem, shape the solution, and make calls that affect real users is still very human. Writing code is only one piece of engineering. Most of the difficult work happens before and after the code. I don’t see a tool doing all of that on its own any time soon. The role of engineers may change, but disappearing completely doesn’t look realistic.