I Interviewed 100 AI Engineers - Here Are the 7 Things They All Regret

Over the past 6 months, I interviewed 100 AI/ML engineers working at:

• Google DeepMind
• Meta AI (FAIR)
• OpenAI
• Anthropic
• Amazon AWS ML
• Smaller AI startups

Asked them one question:

"If you could restart your AI career, what would you do differently?"

Their answers were brutally honest. And surprisingly consistent.

Regret #1: "I Wasted Years Getting the Wrong Degree" (73% Said This)

The most common regret wasn't even close.

73 out of 100 engineers said they regret their educational path.

Google ML Engineer (PhD): "I spent 6 years getting a PhD in theoretical ML. Never used 90% of it. Could've been earning $200K+ for 4 of those years."

Meta AI Researcher (MS): "I did a master's in computer vision. Joined Meta to work on vision. Spent first 2 years doing data pipelines. Should've just started working."

OpenAI Engineer (BS): "Everyone told me I needed a PhD to work in AI. I applied anyway. Got hired. Now I work with 5 PhDs and do the same job. They just have more student debt."

The pattern:

• PhD regretters (41): "Too long, too theoretical, delayed earnings"
• MS regretters (22): "Didn't need it for the job I actually wanted"
• BS regretters (10): "Wished I'd started working sooner instead of more school"

What they wish they'd done instead:

• Start working immediately after BS (49 people)
• Do MS part-time while working (18 people)
• PhD only if going into research (24 people)
• Self-taught with strong portfolio (9 people)

Amazon ML Engineer: "I know people who spent 8 years in school to make $180K. I started after undergrad at $100K. Six years later I'm at $280K. They just finished their PhD last year. My total earnings: $1.2M. Theirs: -$200K in debt."

The math is brutal.

Regret #2: "I Focused on Theory Instead of Building" (68%)

Second most common regret: Learning instead of building.

Anthropic Engineer: "I spent 2 years doing online courses. Watched every Andrew Ng video 3 times. Read every paper on arXiv. Built... zero actual projects."

Google Engineer: "I can explain transformers, attention mechanisms, BERT, GPT architecture in my sleep. Ask me to deploy a model in production? I had no idea when I started."

Startup CTO (former Googler): "Hiring managers don't care if you can derive backpropagation. They care if you can ship features."

What they wish they'd done:

• Build 10 projects instead of taking 10 courses (51 people)
• Contribute to open source earlier (44 people)
• Deploy something to production, even if ugly (38 people)
• Blog about learnings instead of hoarding notes (29 people)

OpenAI Engineer: "My GitHub was empty when I started job hunting. But I had 8 certificates. Guess which one recruiters asked about? Neither. They wanted to see code."

The wake-up call:

One engineer showed me his first AI job interview.

Interviewer: "Walk me through a project."
Him: "I completed Andrew Ng's course and—"
Interviewer: "A project YOU built."
Him: "Well, I followed along with the course labs—"
Interviewer: "Thanks for your time."

Interview lasted 8 minutes.

Regret #3: "I Chased Prestige Instead of Skills" (64%)

Third biggest regret: Optimizing for brand names over learning.

Former Google Engineer: "I turned down a small startup to work at Google. At Google, I spent 3 years optimizing ad CTR. The startup built their own LLM. They're all rich now."

Former Meta Engineer: "Meta looked great on LinkedIn. But I was on the 47-person team for a feature nobody used. Learned nothing. Left after 18 months."

Amazon Engineer: "I took Amazon over a Series A startup because 'FAANG on resume.' Spent 2 years fighting politics instead of building ML systems."

The pattern:

People join prestigious companies for:

• Resume boost
• Salary
• Bragging rights
• Risk aversion

They stay despite:

• Slow promotion tracks
• Bureaucracy killing innovation
• Narrow scope ("I work on 1% of the recommendation engine")
• Boredom

What they wish they'd done:

• Join smaller teams where they'd wear multiple hats (47 people)
• Prioritize learning rate over brand name (38 people)
• Ask "how much will I build?" instead of "what's the salary?" (31 people)
• Join startups earlier in career (29 people)

Ex-OpenAI Engineer (now at startup): "OpenAI was incredible for 18 months. Then it got too big. Now I'm at a 12-person company building our own models. This is where the real learning is."

The contrarian take:

6 engineers said they regret JOINING startups too early.

Google Engineer (ex-startup): "I spent 4 years at failed startups. Learned a ton but made no money. Joined Google at 32. Peers who started at Google at 22 are making $500K. I'm at $220K."

Trade-offs are real.

Regret #4: "I Didn't Specialize Soon Enough" (61%)

Controversial regret: Being a generalist too long.

The conventional wisdom: "Be a generalist early, specialize later."

61 engineers disagree.

Meta Vision ML Lead: "I spent 5 years doing 'general ML.' NLP, vision, recommendation systems, time series. Was decent at all, expert at none. Should've gone deep on vision from day 1."

Google NLP Engineer: "I thought being a jack-of-all-trades made me more employable. Actually made me less valuable. Companies want experts."

OpenAI Researcher: "I finally specialized in language models at year 7. Should've done it at year 1. That's when GPT-2 came out. I'd be leagues ahead now."

Why early specialization matters:

1. Compound learning - 5 years deep > 5 years wide
2. Network effects - Specialists know all the other specialists
3. Higher pay - Companies pay premium for deep expertise
4. Clear positioning - "I'm the vision person" > "I do ML stuff"

What they specialized in:

• Computer Vision (18)
• NLP/LLMs (24)
• Recommendation Systems (11)
• Time Series Forecasting (7)
• Reinforcement Learning (5)
• Speech/Audio ML (4)
• MLOps/Infrastructure (8)

Amazon Recommendations Engineer: "I've been doing recommendation systems for 6 years. Everyone in the industry knows me. Recruiters find me. I don't apply to jobs. Jobs come to me."

The counterargument:

12 engineers said they regret specializing TOO EARLY.

Anthropic Engineer: "I specialized in computer vision. Then transformers came along. Suddenly everyone cares about NLP. Vision jobs dried up. I had to pivot at year 8."

Risk of specialization: Market shifts.

Regret #5: "I Didn't Build My Brand" (59%)

Big surprise: 59% regret not building a public presence earlier.

Google Eng ineer: "I'm a Staff ML Engineer. Great salary. But my peer with the same title has 50K Twitter followers. He gets offered roles I'll never see."

Meta Engineer: "I built incredible systems at Meta. Nobody knows because I never wrote about them. My impact was huge but invisible."

Startup ML Lead: "I spent 7 years heads-down coding. Never blogged, never tweeted, never went to conferences. Then I needed a new job. Zero network. Had to cold apply like a junior."

What "building your brand" means:

• Technical blog (mentioned by 38 people)
• Twitter/LinkedIn presence (mentioned by 31 people)
• Open source contributions (mentioned by 29 people)
• Conference talks (mentioned by 22 people)
• YouTube/courses (mentioned by 11 people)

Engineer with 30K Twitter followers: "My blog posts take 4 hours to write. They've generated $200K+ in opportunities. That's $50K per hour. Best ROI of my career."

Open Source Maintainer: "I maintain a popular ML library. Never applied for a job after that. VCs email me about founding startups. Companies email me with offers. All because my GitHub has stars."

The resistance:

Why don't more engineers do this?

• "I don't have time" (but spend 3 hours/day on Reddit)
• "I have nothing interesting to say" (you know more than 99% of people)
• "I'm not good enough yet" (you'll never feel good enough)
• "It feels like bragging" (it's called marketing yourself)

Ex-Google Engineer, now $1M+ ARR indie maker: "Googlers used to look down on me for tweeting. Now they ask me for advice. Public brand >> prestigious employer."

Regret #6: "I Optimized for Salary, Not Equity" (52%)

More than half regret their comp decisions.

Google Engineer: "I negotiated hard for $220K salary at Google. Turned down startup offering $150K + 0.5% equity. That startup sold for $800M. I left $4M on the table."

Amazon Engineer: "Amazon offered $180K all cash. Startup offered $120K + 1% equity. I took Amazon for 'stability.' That startup is now worth $2B. Missed $20M."

Meta Engineer: "I watched 5 people leave Meta for startups. Thought they were crazy. All 5 made more money than me in 4 years despite my promotions."

The pattern:

Early career (years 0-3): Take highest salary

• Need to pay off debt
• Build emergency fund
• Establish baseline

Mid career (years 4-10): Consider equity

• Financial stability established
• Can take calculated risks
• Asymmetric upside potential

What they wish they'd done:

• Join 1-2 startups before 30 (39 people)
• Take less salary for meaningful equity (31 people)
• Understand equity before negotiating (28 people)
• Ask "what's the upside?" not just "what's the salary?" (22 people)

Engineer who left Google for startup: "At Google I'd make $300K this year, $320K next year, $340K the year after. Predictable. Capped. At my startup, I might make $150K for 3 years then $5M. Or $0. But the upside is real."

The counterpoint:

18 engineers said they regret taking too much risk.

3-time failed startup employee: "I chased equity for 8 years. 3 startups. All failed. Made $100K total in 8 years. My Google friends are millionaires. I have nothing."

Risk tolerance matters.

Regret #7: "I Didn't Change Jobs Enough" (47%)

Last major regret: Staying too long.

7-year Meta employee: "I stayed at Meta because it was comfortable. Colleagues who left every 2-3 years are making 50% more than me."

Google lifer: "I'm at Google 9 years. Love it. But my salary has increased 40% total. People who job-hop every 2 years have doubled their comp."

Amazon 5-year vet: "I keep getting 3% raises. Recruiter offered me $100K more to leave. Amazon won't match. If I'd left 2 years ago and come back, I'd make $150K more."

The job-hopping advantage:

• Year 1-2: First company, $100K
• Year 3-4: New company, $140K (40% jump)
• Year 5-6: Another jump, $190K (36% jump)
• Year 7-8: Next company, $250K (32% jump)

vs staying:

• Year 1-2: $100K
• Year 3-4: $110K (3% annual raises)
• Year 5-6: $120K (3% annual raises)
• Year 7-8: $132K (3% annual raises)

Difference: $118K per year by year 8.

What they wish they'd done:

• Change jobs every 2-3 years (37 people)
• Interview even when happy (31 people)
• Always know market rate (26 people)
• Leave when growth plateaus (22 people)

Serial job-hopper (5 companies in 10 years): "Each jump was 25-40% raise. Total comp went from $80K to $320K. My friend who stayed at one company: $80K to $150K."

The loyalty trap:

Companies reward hopping, not loyalty.

The Common Thread: They All Wish They'd Moved Faster

Across all 100 interviews, one theme emerged:

"I wish I'd moved faster."

• Faster to start working (not more school)
• Faster to build (not learn)
• Faster to specialize (not generalize)
• Faster to build brand (not hide)
• Faster to take equity (not optimize salary)
• Faster to change jobs (not stay comfortable)

OpenAI Engineer: "I spent my 20s preparing to do cool work. Should've just started doing cool work. The preparation never ends."

Google Researcher: "I told myself 'I'll build my Twitter presence after I'm more senior.' Now I'm senior and starting from zero. Should've started at day 1."

Meta Engineer: "I waited until I was 'good enough' to contribute to open source. Spoiler: You're never ready. Just start."

What You Should Do Differently

Based on 100 interviews, here's the playbook:

Years 0-2: Build + Ship

• Skip extra degrees (unless targeting research)
• Build 10 projects, deploy 3
• Contribute to open source
• Start technical blog
• Join company with strong ML team

Years 3-5: Specialize + Brand

• Pick your specialty (vision, NLP, etc.)
• Go deep, become known for it
• Write, speak, teach publicly
• Build network in your specialty
• Consider first startup for equity

Years 6-10: Leverage + Multiply

• You're now an "expert"
• Negotiate from position of strength
• Take calculated equity bets
• Build products/companies
• Help others (consulting, advising)

The meta-lesson:

Nobody regrets moving too fast.

Everyone regrets waiting.

Frequently Asked Questions

Should I get a PhD for AI?

Only if you want to do research at top labs. Otherwise, no. 73% of engineers regretted their extended education.

How many projects should I build?

Start with 3 solid projects. Better than 10 mediocre tutorials. Focus on deployment, not toy examples.

When should I specialize?

Earlier than you think. Years 1-2 instead of years 5-7. Compound learning is real.

Do I really need a public brand?

You can succeed without it. But 59% of successful engineers wish they'd built one earlier. Asymmetric upside.

Should I take startup equity over FAANG salary?

Depends on your risk tolerance and financial situation. But 52% of engineers regret choosing only salary.

How often should I change jobs?

Every 2-3 years if growth plateaus. Loyalty doesn't pay like job hopping does.

---

The Brutal Truth

After 100 interviews, the pattern is clear:

The AI engineers who moved fast, built in public, specialized early, and took calculated risks are the ones with the most options and highest comp.

The ones who played it safe, stayed quiet, generalized, and optimized for stability have comfortable careers.

But they all have regrets.

Which path will you choose?

---

Learn from those who've been there:

• AI Career Quiz - Find your optimal path
• AI Salary Calculator - Know your market rate
• AI Career Resources - Weekly insights from industry leaders

This article is based on 100 interviews conducted over 6 months with AI/ML engineers at major tech companies and startups.