iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Learning to code with AI did not make Chetan Vashistth an expert overnight. In his first-person account, the turning point was a small, personal project: turning a hand-drawn webpage sketch into HTML and CSS with ChatGPT. The rest of the journey was slower—learning new tools, getting comfortable in the terminal, recovering from a database mistake, and returning to software fundamentals.
How the first project made coding feel possible
Vashistth describes early experiments with ChatGPT as exciting, but the sketch-to-webpage task gave that excitement a practical shape. He had an idea he could see on paper and used AI to help turn it into a working page. That experience made starting feel more approachable; it does not prove that AI is a better teacher than a course or that the same approach works for every beginner.
The value of a project like this is that it gives a learner something concrete to question. Instead of asking an AI to teach “programming” in the abstract, you can ask what a particular line does, why a layout behaves a certain way, or what to change to get a visible result. The page is a starting point for learning, not evidence that the learner understands the code.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteWhy tool fluency took longer than expected
Vashistth’s account moves from ChatGPT to Cursor and Claude Code, but adopting a new tool did not remove the ordinary friction of development. He recalls struggling with copying and pasting code, then spending days refining a terminal workflow. Learning how a tool interacts with files, a repository, and a command line is part of the work—not a sign that you are failing to learn programming.
His story also includes a database incident that pushed him toward stronger configuration and guardrails. The practical lesson is to treat AI-generated changes as real changes to a real project: understand which files and systems are affected, test carefully, and avoid giving a tool unchecked access to data or environments you cannot afford to damage. The account does not establish a universal setup or a specific safe configuration for every project.
What studies report about speed and understanding
AI can help people finish a coding task faster without ensuring they understand the solution. Two studies illustrate why those outcomes should be kept separate rather than collapsed into a single claim that AI either improves or harms learning.
| Study | Participants and task | Reported result | What it does not establish |
|---|---|---|---|
| ACM ICER, 2025 | 10 undergraduate computing students completed brownfield tasks in an unfamiliar legacy web app, with and without Copilot. | The authors reported that participants completed tasks 34.9% faster with Copilot and made 50% more solution progress. Interviews also raised concerns about understanding why suggestions worked. | This small, task-specific experiment does not establish a general productivity or learning effect for all programmers, projects, or AI tools. |
| Anthropic, January 29, 2026 | 52 mostly junior software engineers who knew Python but were unfamiliar with the Trio library were randomized to AI-assisted or hand-coding learning. | The AI-assisted group averaged 50% on an immediate quiz; the hand-coding group averaged 67%. The report describes the difference as statistically significant. | This measured immediate comprehension of a new library, not long-term skill or absolute beginners learning to program from scratch. |
These results concern different people, tasks, and outcomes; they are not a head-to-head comparison. The Copilot experiment examined task completion on legacy code, while Anthropic examined immediate comprehension after learning an unfamiliar Python library. Neither shows that AI inevitably prevents learning, and neither identifies a best AI product.
Course-based research offers a complementary point. A 2024 IEEE study integrated ChatGPT and Copilot into introductory programming activities that emphasized critical thinking. Its abstract reports increased student awareness of AI’s possibilities and limitations, alongside increased reported critical-thinking practices. That is evidence from a course assignment, not a guarantee about every learner or tool.
How to use AI without handing over the learning
Anthropic’s analysis found an association between stronger quiz performance and asking AI for explanations or conceptual guidance rather than simply delegating code. The authors caution that this qualitative grouping does not establish causation. It is still a useful way to structure practice: keep yourself responsible for understanding and checking the result.
- Start with a goal you can inspect. Pick a small feature, such as changing a page layout or adding a simple interaction, and decide what a correct result should do.
- Ask for reasoning, not just output. When AI proposes code, ask what each unfamiliar part does, why that approach fits the problem, and what alternatives or assumptions matter.
- Read the change before accepting it. Look at the affected files and compare the proposed code with what was there before. If you cannot explain a line, pause and investigate it.
- Run the code and test the behavior. Check the result against your goal, including likely edge cases. A plausible-looking answer is not the same as a working, correct change.
- Practice debugging deliberately. When something breaks, inspect the error and trace the relevant code before asking for a fix. Then verify the proposed fix and explain what caused the original problem.
The point is not to reject useful assistance. It is to avoid confusing a finished task with a skill you can use again. A small study of students coding in an unfamiliar legacy app found faster completion with Copilot, while interviews surfaced concerns about understanding suggestions; Anthropic’s separate trial found lower immediate quiz results among AI-assisted learners. Together, the evidence makes a case for checking comprehension, not for a blanket rule against AI.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why returning to fundamentals mattered
Vashistth describes experimenting with MCP and Blender before returning to software design fundamentals and core books. That return is not a rejection of new tools. It reflects a practical distinction: a tool can help produce an output, while programming knowledge helps you decide what to build, judge whether the output is sound, and diagnose what happens when it fails.
For a learner, fundamentals need not mean postponing projects until every concept is mastered. They can be learned alongside a project: use AI to clarify a concept, then apply it, inspect the result, and revisit the underlying idea when the code surprises you. Vashistth closes his account with a modest measure of progress: “That is where I am today. Still figuring it out — just faster than before.”
What this two-year account can—and cannot—tell you
This is one developer’s account of changing tools, frustrations, mistakes, and learning habits. It is useful as a realistic example of gradual progress, not as a measured outcome for beginners generally. The evidence here does not provide a population-level statistic for how quickly people learn to code with AI. A sound expectation is narrower: AI can help make a project feel approachable and accelerate some tasks, but building skill still depends on understanding, testing, and debugging the code you use.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
