Dev Chroniclesby Thinnd
Conversations

A Fresher and a Senior Talk About Learning the Craft in the AI Era

A rebuilt coffee chat between Mai, a worried fresher, and Khoa, a senior. They speak plainly about fear of losing jobs, letting AI do the exercises, what to learn, who to ask, how seniors use AI and what employers look at, with evidence from studies.

Nhu Duy Thin (Thinnd)8 min read
  • #fresher
  • #senior
  • #AI
  • #learning
  • #mentoring
  • #dialogue

Slide deck

Click the slides, then use the arrow keys to move. Press L to change language and F for full screen.

Open full screen

For people new to the field, and for seniors guiding them.

Mai and Khoa are rebuilt characters to make this easy to picture, not real people. The conversation is an illustration, and the figures come from the public studies listed at the end. I say which parts are data and which are the characters' opinions. Many studies are small or published by vendors, so read them with care.

Quick summary#

  • Nobody can promise the market will not change. But understanding, debugging and explaining code are still what hiring tests.
  • Letting AI write your exercises can leave you learning less. Write first, ask AI to explain afterward.
  • Ask your senior after trying yourself, and bring what you tried.
  • Seniors must verify AI too. Feeling fast is not a measurement.
  • Each month ask yourself three questions to see whether you are really progressing.

Setting#

A Friday afternoon at a café near the office. Mai has been at the company three months, on probation. Khoa is a senior on Mai's team, eight years in the field.

1. "I am afraid AI will take my job"#

Mai: Khoa, honestly, I am really worried. A friend says companies are hiring fewer freshers because AI can write code now.

Khoa: I cannot promise you anything firm, because I do not know. But let me share what I have read. A Stanford study using payroll data found employment of 22 to 25 year olds in the most AI-exposed occupations fell about 13% relative to other groups.

Mai: So it is true, then?

Khoa: Hold on. That is American data, and it is a correlation, meaning two things moved together, not proof that one caused the other. Press coverage often names programming specifically, but the paper's summary does not single out that job. I have not seen an equivalent figure for Vietnam. So my view is: moderate worry is reasonable, just do not let it decide how you learn.

Takeaway: the door may be narrower, but the data is only a correlation and from elsewhere. Focus on what you control: your own skills.

2. "I let AI do all my exercises, it is so fast"#

Mai: There is something I am a bit embarrassed about. I often paste the task into AI, it writes code that works right away, and I submit. It is so fast.

Khoa: Remember my review of your code last week? I asked why you used nested loops and you could not explain.

Mai: Yes... I only knew it ran.

Khoa: Let me tell you about an Anthropic trial published in early 2026. They had 52 programmers, mostly early-career, learn an unfamiliar Python library. The AI group then scored about 50% on a knowledge quiz and the hand-coding group about 67%. The biggest gap was on debugging questions.

Mai: So AI makes me dumber?

Khoa: I would not say that for sure. The sample is small, 26 per group, and I have read the summary, not the whole paper. The AI group was also about two minutes faster, but that difference was not statistically significant. Still, it matches what I observe: people who let AI write the very part they need to learn end up short on debugging.

Mai: So what should I do?

Khoa: Try this rule: write for 20 minutes first, then ask AI to explain or point out your mistakes. Do not ask for new code. Ask for an explanation.

Takeaway: AI is better at explaining than at learning for you. Write first so you have a version of your own to compare.

3. "So what should I learn?"#

Mai: If AI can write simple code, what should I learn?

Khoa: I think five things, and to be clear this is my opinion, not a verified fact:

  1. Read other people's code. Each week read an unfamiliar file and say in words what it does.
  2. Debug systematically. Reproduce, narrow, hypothesize, test. Do not guess.
  3. Write tests. Tests tell you what "correct" means.
  4. Use Git and reviews. Working with others is daily life.
  5. Explain decisions. Why this way and not that.

Mai: Those are all things AI can do too, no?

Khoa: AI does part of it, but the person who answers for the result needs to understand. Read the post on a 12-month plan for freshers, it goes into more detail than I do.

4. "Should I ask AI or you?"#

Mai: Sometimes I do not know whether to ask AI or bother you.

Khoa: I split it like this. Ask AI about concepts and examples: "Explain pointers", "give an example of a loop". Ask me about context and decisions: "Why does our project do it this way", "should this part be changed".

Mai: And how should I ask you?

Khoa: Try for about 15 minutes first. Then bring three things: what you tried, what you expected, and what you saw. That lets me answer fast, and I can see how you think.

Mai: But I am afraid of asking something stupid...

Khoa: A question with those three things is never stupid. The stupid one is where you did not try first and cannot say what you tried.

5. "And how do you use AI?"#

Mai: Do you use AI?

Khoa: Yes, daily. For repetitive parts such as scaffolding, sample tests and explaining odd code. But I still read every change and keep a checklist for reviewing. You can look at the 15-point checklist I often use.

Mai: So you must be much faster?

Khoa: I thought so, then I read METR's study from July 2025. They had 16 experienced developers do 246 real tasks, half allowed AI and half not. The AI group was about 19% slower, yet they themselves believed they were about 20% faster.

Mai: Huh, so AI does not help at all?

Khoa: Careful, I am not saying that. The sample is small, they worked in projects they knew very well, and the tools were early 2025, much changed since. I draw only one lesson: feeling fast is not a measurement. So I started timing a few tasks myself, with and without AI.

Takeaway: even seniors need to verify. Measure instead of trusting a feeling.

6. "How do I know I am improving?"#

Mai: I do not know whether I am really improving.

Khoa: Each month ask yourself three questions and write them on paper:

  1. What can I do this month that I could not last month (without AI)?
  2. Of the code I submitted, how much can I explain without looking back?
  3. How did I handle the hardest bug this month, how long did it take, who helped?

If questions one and two do not move for two months, reduce AI writing and increase hand-writing. If question three is always "asked AI to fix it", practice debugging by hand.

7. "What do they look at in interviews?"#

Mai: Later, in interviews, what do they look at in me?

Khoa: I have interviewed a few freshers. I look at three things: can you explain your own code, can you debug in front of me, and how do you verify what AI gives you. The Stack Overflow 2025 survey, with more than 49,000 respondents, found 46% do not trust AI's accuracy and 66% often meet "almost right but not quite" answers. So I care how you verify more than how much AI you use.

Mai: Do you ask "how do you use AI"?

Khoa: Yes. A good answer usually has three parts: what you use it for, how you verify, and when you do not use it.

8. As the conversation winds down#

Mai: Give me some homework.

Khoa: This week: pick a small task, write it yourself for 20 minutes, then ask AI to explain where you went wrong. At the weekend, tell me what you learned.

Mai: Are you worried about anything?

Khoa: Yes, a few things: the data is still thin, and I am not sure which parts will hold up long term. I am still learning too. Do not take what I say as truth. Take it as a starting point for you to try and verify yourself.

Takeaways#

  1. Worry moderately, and do not let it decide how you learn. The jobs data is a correlation from elsewhere.
  2. Write first, ask AI to explain afterward. The small Anthropic trial suggests letting AI write leaves a gap in debugging.
  3. Ask people after trying yourself, and bring what you tried.
  4. Verify and measure, even as a senior. METR's study is a reminder that feeling fast is not a measurement.
  5. Ask yourself three questions each month to know whether you are really improving or just feeling fast.

A balanced view#

  • This is a rebuilt dialogue. Khoa's views are a character's, not verified conclusions.
  • The studies mentioned are small or published by vendors. The jobs data is American. Both may change as tools and the market change.
  • Everyone learns differently. Use these suggestions as a starting point, not hard rules.

References#

Frequently asked questions

Should a fresher worry that AI will take their job?
Moderate worry is reasonable, but do not let it decide how you learn. A Stanford study found employment of 22 to 25 year olds in the most AI-exposed occupations fell about 13% relative to other groups, but it is American data and a correlation that does not prove cause. More certain is that understanding the problem, debugging and explaining code are still tested in hiring.
Should I ask AI or a senior?
Ask AI about concepts and examples, ask a senior about context and project decisions. Before asking a person, try for about 15 minutes and bring what you tried, what you expected and what you saw. That shows the senior how you think and helps them answer the right thing.
How do seniors use AI at work?
Typically for repetitive parts such as scaffolding, sample tests and explaining unfamiliar code, while still reading and verifying the result. METR's study of 16 experienced developers found them about 19% slower with early-2025 AI tools even though they believed they were faster, so many seniors choose to time themselves rather than trust a feeling.
What do employers look at in an AI-era fresher?
Whether you can explain your own code, debug live and verify AI output. The Stack Overflow 2025 survey found 46% of users do not trust AI's accuracy, so interviewers care how you verify more than how much AI you use.