🎧 Listen to this article
The Best Time to Build with AI
‘It really feels like th" cover: image: "" alt: “AI in Business News” hidden: true
Based on Stanford CS230 lecture featuring Andrew Ng and Lawrence Morony on career development in AI.
Part 1: Andrew Ng’s Perspective
The Best Time to Build with AI
“It really feels like the best opportunity, the best time ever to be building with AI and building a career in AI.”
Evidence of Progress:
- METR study: Task complexity AI can handle doubles every 7 months
- AI coding: Doubling time even shorter (~70 days)
- Models can do increasingly complex tasks as measured by human completion time
Two Themes: More Powerful and Faster
| Theme | What It Means |
|---|---|
| More Powerful | AI building blocks (LLMs, RAG, agentic workflows, voice AI, deep learning) let you build software more powerful than anyone could a year ago |
| Faster | AI coding tools dramatically speed up development |
AI Building Blocks:
- Large language models
- RAG (Retrieval-Augmented Generation)
- Agentic workflows
- Voice AI
- Deep learning (LLMs have decent understanding of this too)
Stay on the Frontier of Tools
“Being half a generation behind in tools means being quite a bit less productive.”
- Tools change rapidly (every 3-6 months)
- Current favorites mentioned: Claude Code, OpenAI Codex, Gemini
- Staying current = higher productivity
The Product Management Bottleneck
Key insight: When code generation becomes easy, the bottleneck shifts to deciding what to build.
The Build-Feedback Loop:
- Write software
- Show to users, get feedback
- Revise conception (UI issues, feature requests)
- Iterate
Changing ratios:
- Traditional: 1 PM : 4-8 engineers
- Emerging: 1 PM : 1-2 engineers (or engineer + PM combined)
Career advice: Engineers who can talk to users and develop empathy for what to build move fastest.
The People You Work With Matter Most
“One of the strongest predictors for your speed of learning and level of success is the people you surround yourself with.”
Stanford advantage: Rich “connective tissue” to frontier AI labs - former students, friendships, inside knowledge.
Cautionary story: Stanford student joined company with “hot AI brand” without knowing team assignment. Ended up on Java payment processing backend, not AI. Left frustrated after a year.
Advice:
- Find out who you’ll actually work with day-to-day
- A good team at a less famous company > hot brand with unknown team
- You learn from people, not company logos
Go Build Stuff
“The number of ideas out in the world is much greater than the number of people with the skill to build them.”
- Don’t wait for permission
- Cost of failure is low (waste a weekend, learn something)
- Be responsible, don’t harm others
- Build things that no one else will build
Work Hard
“All of my PhD students that became very successful, I saw every single one of them work incredibly hard.”
Politically incorrect but true:
- If you’re fortunate enough to be in a position to work hard, do it
- 2 a.m. debugging sessions, high-performance tuning - still doing it
- Between watching TV vs. building with agentic coder - choose building
- Respect those who can’t work hard (injury, disability, life circumstances)
Part 2: Lawrence Morony’s Market Perspective
Job Market Reality Check
Current signals:
- Junior hiring slowing significantly
- High-profile layoffs dominating headlines
- Entry-level positions feel scarce
- Competition is fierce
But should you worry? No.
“People with the right mindset will thrive.”
The AI Industry Evolution
| Period | What Happened |
|---|---|
| Pre-2020 | AI winter, limited adoption |
| 2020-2021 | Global pandemic, industrial slowdown |
| 2022-2023 | Post-COVID hiring surge + AI explosion = massive overhiring |
| 2024-2025 | “Great wake-up” - correction, underqualified people laid off |
Key insight: Companies overhired people with “AI on resume” without proper vetting. Now they’re more cautious.
Three Pillars of Success
| Pillar | Description |
|---|---|
| Understanding in Depth | Academic knowledge (read papers, understand architectures) + pulse on trends (signal vs. noise) |
| Business Focus | Align output with business needs. “Don’t let your output be for the job you have, let it be for the job you want.” |
| Bias Toward Delivery | Ideas are cheap, execution is everything. Ground your ideas. |
What Working in AI Looks Like Now
Then (2-3 years ago): Build an image classifier, get six figures thrown at you.
Now: Everything is about production.
- What can you do for production?
- Optimizing models
- Understanding users, UX
- Driving business value
Four realities:
- Business focus is non-negotiable - The pendulum swung back from “bring your whole self to work” activism to business-first
- Risk mitigation is part of the job - Understand risks of AI transformation, help mitigate them
- Responsibility is evolving - From fluffy “AI works for everybody” to “AI works, drives business, then works for everybody”
- Learning from mistakes is constant - Give grace to others who make mistakes
Responsible AI Lessons (Gemini Example)
What went wrong:
- Safety filters blocked “Caucasian” and “white” prompts but allowed other ethnicities
- Generated Irish people always with red hair (8% reality, 100% in output)
- Naive solutions caused reputation damage
Lesson: Responsible AI has moved from social issues to hard business requirements - preventing reputation damage.
Technical Debt and Vibe Coding
The Technical Debt Framework
“Every time you build something, you take on debt.”
Debt includes:
- Bugs to fix
- People to convince
- Documentation to write
- Features to add
- Support to provide
The only way to avoid debt: Do nothing.
Good Debt vs. Bad Debt
| Good Debt (Mortgage) | Bad Debt (Credit Card) |
|---|---|
| Clear objectives met | Solution looking for problem |
| Business value delivered | Spaghetti code |
| Human understanding | Authority over merit (VP’s bad code) |
Key insight: The more skilled you are as an engineer, the better you become at vibe coding. Engineers understand implications of generated code best.
Managing Technical Debt
Good debt checklist:
- Clear objectives - what are you building and why?
- Business value delivered - does it help the business?
- Human understanding - can others maintain this code?
Bad debt warning signs:
- Solution looking for a problem
- Spaghetti code from repeated prompting
- Platform mismatch (e.g., iOS code for macOS app)
- Authority over merit (non-technical VP’s vibe-coded mess)
Navigating the Hype Cycle
The Anatomy of Hype
“The currency of social media is engagement. Accuracy is not the currency.”
Hype pyramid (top to bottom):
- Hype (engagement farming)
- Massive VC investment (already drying up)
- Unrealistic valuations
- Me-too products
- Real value (small kernel at bottom)
Becoming a Trusted Advisor
When someone says “help me implement an agent”:
- Ask “Why?” - Peel apart the real need
- Ask “What do you want to do?” - Get specific
- Then discuss AI solutions
Example: CEO wanted “agents” to save costs. After peeling: wanted salespeople to be more efficient. Solution: Research automation tool, not generic “agent.”
The Agentic Framework
| Step | Description |
|---|---|
| 1. Understand Intent | Use LLM to understand the task and context |
| 2. Plan | Declare available tools, let LLM break into steps |
| 3. Use Tools | Execute the plan with specified tools |
| 4. Reflect | Check if intent was met, loop if needed |
Example application: Video generation went from hallucinating crowds in empty hockey rink to emotionally nuanced movie scenes by using agentic workflow.
Hype Navigation Strategy
- Filter actively - Recognize engagement farming
- Go deep on fundamentals - Understand how things actually work
- Keep finger on pulse - Stay connected to trends without drowning in noise
- Make it mundane - When you can explain something in boring detail, you truly understand it
Example: Text-to-video isn’t magic - it’s predicting successive frames based on training data. That mundane understanding lets you build better solutions.
The AI Industry Bifurcation
Big AI vs. Small AI
| Big AI | Small AI |
|---|---|
| Hosted by others (GPT, Gemini, Claude) | Self-hosted models |
| Drive toward AGI | Fine-tuned for specific tasks |
| Sooner bubble | Later bubble |
| High cost, API dependency | Privacy, control, lower cost |
Key trend: Open-weights / self-hostable models exploding. 80% of Y Combinator companies using small models (many from China).
Why small AI matters:
- Privacy (Hollywood IP, medical records, legal documents)
- Latency (on-device processing)
- Cost (no cloud service to stand up)
- Control (fine-tune for downstream tasks)
Skills needed: Fine-tuning, understanding model architectures, deploying self-hosted models.
Embedded Intelligence Everywhere
Hardware evolution:
- CPU + GPU model changing
- SME (Scalable Matrix Extensions) - AI workloads on CPU
- Apple’s neural cores, Chinese phone vendors (Vivo, Oppo)
Example: Alipay app doing on-device photo search - no cloud, no latency, no privacy concerns.
Interview and Career Advice
The “10x Engineer” Trap
Story: Talented coder failed 300+ interviews. Why?
- Advice to “stand your ground” made him hostile in interviews
- Being right != being hireable
- Companies choose people they want to work with
Lesson: Technical skill + teamwork attitude = success.
Produce Output for the Job You Want
Lawrence’s Google story:
- Failed twice interviewing as PM
- Third time: Built a Java app running on Google Cloud, put code on resume
- Entire interview became discussion of his code
- Power shifted to him - talking about what he knew
Diversify Your Skills
“Don’t be that one-trick pony who only knows how to do one thing.”
Diversification means:
- Model knowledge (LLMs, computer vision, etc.)
- Application building
- Scaling and software engineering
- User experience
- Business understanding
Even for narrow job requirements: Show skills in multiple areas.
Key Stories and Examples
The Non-Technical Success Story
Former hockey player:
- Dropped out of school at 13
- Runs nonprofit ice rinks
- Used ChatGPT to consolidate data from spreadsheets, PDFs, machines
- Saved $150K/year in consulting fees
- Money now goes to underprivileged kids’ hockey equipment
“Congratulations, you’re now a developer.”
The Brain Cancer Researcher
Problem: 10 researchers sharing 1 GPU. Half-day per week each.
Solution: Google Colab - free GPU in cloud. Changed everything.
The TensorFlow Certificate Story
- Program cost Google $100-150K/year
- Deliberately revenue-neutral
- Man lifted from Syria war zone to German tech job
- Google cancelled it due to no revenue
Lesson: Sometimes good programs die for business reasons.
Final Principles
Assume Good Intent, Prepare for Bad
“Any tool can be used for any means. Educate and inspire people towards using things for the correct means.”
- AI itself has no choice in how it’s used
- Most mistakes are good intent applied wrongly
- Governance can cause more problems than it solves
The Trusted Advisor Mindset
Your role:
- Filter signal from noise for others
- Explain technical reality to leadership
- Make complex things mundane and understandable
- Help non-technical people succeed with AI
Navigate Bubbles Successfully
Dot-com bubble lesson: Companies that did it right (Amazon, Google) not only survived but thrived.
AI bubble coming: Those who focus on fundamentals, build real solutions, understand business side, and diversify skills will thrive through and after.
Summary Checklist
For Job Seekers
- Know who you’ll work with day-to-day
- Produce output for the job you want (build and show)
- Demonstrate business understanding
- Show you can deliver, not just ideate
- Be someone people want to work with
For Career Development
- Stay on frontier of AI coding tools
- Build things - don’t wait for permission
- Develop product sense (what to build)
- Surround yourself with inspiring people
- Work hard if you’re able to
For Being a Trusted Advisor
- Ask “why?” before discussing solutions
- Filter hype from signal
- Make technical concepts mundane and explainable
- Understand business needs first, then apply AI
- Manage technical debt responsibly
Key Quotes
“The best time ever to be building with AI.”
“Being half a generation behind in tools means being quite a bit less productive.”
“The bottleneck increasingly is deciding what to build.”
“You learn from the people you deal with day-to-day, not the company logo.”
“The number of ideas is much greater than the number of people with the skill to build them.”
“Don’t let your output be for the job you have, let your output be for the job you want.”
“Ideas are cheap, execution is everything.”
“The currency of social media is engagement. Accuracy is not the currency.”
“When you can make something mundane, you truly understand it.”
“Assume good intent, but prepare for bad intent.”
Based on Stanford CS230 lecture. Features Andrew Ng (Stanford, DeepLearning.AI) and Lawrence Morony (ARM, former Google AI Advocate, author of multiple AI books including TensorFlow and PyTorch guides).
Disclaimer: This blog post was automatically generated using AI technology based on news summaries. The information provided is for general informational purposes only and should not be considered as professional advice or an official statement. Facts and events mentioned have not been independently verified. Readers should conduct their own research before making any decisions based on this content. We do not guarantee the accuracy, completeness, or reliability of the information presented.