Silicon Valley GirlGitHub CEO: Why Now Is the BEST Time to Be a Developer | Thomas Dohmke
CHAPTERS
- 0:50 – 1:53
Vibe coding explained: letting agents drive inside your IDE
Thomas defines “vibe coding” as working in an IDE’s agent mode (Copilot/Cursor/etc.) where you mostly interact with an AI agent instead of writing and reviewing code line-by-line. Marina probes whether this means you can build software without learning to code, and Thomas clarifies what’s changing in the workflow.
- •Definition of vibe coding: assign tasks to an agent and follow its proposed steps
- •Reduced code review/understanding in the loop compared to traditional development
- •Examples of tools enabling the workflow (Copilot, Cursor, Windsurf)
- •Shift from writing code to directing and validating an agent
- 1:53 – 3:35
How far can vibe coding go before you hit a wall?
Marina asks how complex a vibe-coded product can get (databases, auth, scaling). Thomas argues the ceiling is largely determined by your prompting skill and patience, but real complexity eventually forces you to understand the code—especially when performance and scalability matter.
- •Complexity is bounded by prompting ability when you don’t understand the code
- •Analogy to image generation: iterative prompting works until it doesn’t
- •You can build common features (pages, auth, settings) with agents
- •Scaling/performance (e.g., Black Friday load) still demands professional engineering
- 3:35 – 4:48
“No developers in 2 years?” Why coding skill still differentiates businesses
Thomas rejects the idea that AI lets non-coders create a defensible billion-dollar tech business effortlessly. If everyone can prompt the same output, differentiation disappears; the winners will build much more complex, AI-accelerated products that still require developer expertise.
- •If AI makes building trivial for everyone, it eliminates moat and differentiation
- •Tech businesses still need developers to build unique, complex systems
- •AI enables startups to build 10x–100x more complex products, not just faster MVPs
- •Many non-coding businesses exist, but tech advantage still comes from building
- 4:48 – 6:10
Prompting as a superpower: AI as your co-founder (sponsor segment)
Marina frames AI as a potential “co-founder,” emphasizing that outcomes depend on how well you prompt and structure instructions. She highlights a HubSpot prompt-engineering guide and then returns to the core question: will the world need more or fewer developers?
- •AI value depends on communication quality, not just access to the tool
- •Systematic prompting frameworks vs. “random prompting”
- •Building reusable prompt components to save time
- •Transition back to developer-demand discussion
- 6:10 – 7:15
Why there will be more developers: democratized learning and new tiers of builders
Thomas predicts “way more developers” because AI removes early learning friction and helps people unblock faster. He distinguishes between casual “consumer developers” building personal micro-apps and professional developers building advanced systems and agents.
- •AI lowers barriers for kids and beginners to start building immediately
- •Agents help learners debug and overcome being stuck without local support
- •Emergence of a spectrum: consumer micro-app builders → professional engineers
- •Professional developers remain crucial for building AI systems and infrastructure
- 7:15 – 8:24
The smartest companies will hire more developers: 10x devs and 100x impact
Thomas argues AI should be used to accelerate output rather than reduce headcount. If each developer becomes dramatically more productive, strong companies will scale teams to multiply results—similar to how modern website builders didn’t eliminate the need for specialists.
- •AI-driven productivity compounding: 10x a dev → 10 devs can do 100x
- •Parallel to Squarespace-era shift: more creators, continued need for experts
- •VC perspective: undifferentiated “simple website” businesses have no moat
- •Best strategy: grow capability and ambition, not just cost savings
- 8:24 – 9:45
“Who’s buying?” Backlogs grow, not shrink—and 90% agent-written code can still mean more work
Marina raises the demand-side concern: if teams ship more, will there be enough customers? Thomas says this is a short-term worry; in practice AI expands backlogs and ambition, and even if agents write most code, total code volume can grow enough to keep developers fully utilized.
- •Short-term market uncertainty can make increased output feel demand-limited
- •AI tends to add work by expanding what’s feasible and desirable
- •Prediction: ~90% of code written by agents
- •If total code grows 10x, developers may still do the same absolute amount (or more)
- 9:45 – 10:59
Why big tech pauses hiring: uncertainty, transition, and mandates to use AI
Thomas interprets hiring slowdowns as a response to rapid change and uncertainty rather than a permanent reduction in developer need. Companies are evaluating who can thrive in an AI-first environment, and leaders are investing heavily to stay competitive.
- •Hiring pauses reflect market and political uncertainty and risk management
- •AI adoption becomes a competitive necessity; refusal to use AI is a liability
- •This period is framed as a transition phase before re-acceleration
- •Example of aggressive investment as signal of direction (Scale AI acquisition mention)
- 10:59 – 12:24
Advice for new coders: learn with AI and leverage the “teen advantage”
Thomas advises learners to adopt AI tools from the start, arguing younger people have more time and openness to experiment than busy professionals. He predicts the next generation will treat AI agents as a default layer across tasks, including software development.
- •Core advice: learn coding with AI, not in spite of it
- •Young learners can iterate faster due to time and openness
- •Analogy: growing up with AI like Gen Z grew up with smartphones
- •Agents will become standard across work (email, planning, travel, coding)
- 12:24 – 14:19
Best platforms to start vibe coding—and why you still get stuck
Asked whether it’s the right time to try vibe coding, Thomas says yes: many tools now let beginners build without deep technical setup. He cautions that most people hit limits—either the app is too simple/ugly or they reach the edge and must edit source code and understand deployment basics.
- •It’s a good time: tools like ChatGPT/Claude and OpenAI Codex enable building
- •Some tools still require basics (repos, where code goes, deployment)
- •Beginner-friendly builders mentioned: Vercel, Lovable, Bolt, etc.
- •Common failure modes: not knowing what to ask, shallow prompts, quality issues, edge-case limits
- 14:19 – 15:10
Will AI have better ideas than humans? AI as a thinking partner
Thomas argues AI can improve idea generation by helping with reflection, combining notes, and exploring alternatives—but human motivation, excitement, and “what keeps you up at night” remain key inputs. AI then helps expand and operationalize those human-originated sparks.
- •AI supports reflection: “what am I missing?” and combination thinking
- •Reasoning capabilities amplify human triggers and instincts
- •AI helps turn ideas into artifacts (e.g., pitch decks)
- •Human origin of passion and direction remains central
- 15:10 – 16:43
AGI by 2030? Definitions, capability vs. creativity, and the emotion gap
Thomas says AGI timelines depend on definitions: models may already exceed humans in knowledge and summarization, but lack creativity rooted in emotion and sentience. He notes that some technologies (self-driving, vibe coding) can feel “AGI-like,” yet true human-like instinct and novelty are not imminent.
- •AGI is definition-dependent; debate exists on both sides
- •Models surpass humans in stored knowledge and rapid summarization
- •Key missing pieces: creativity, emotion, sentience
- •Examples that feel AGI-like (Waymo, vibe coding) but don’t equal human-level instinct
- 16:43 – 19:04
No fear—excitement: what Thomas teaches his kids in an AI era
Thomas explains why he’s optimistic: we’re living through a uniquely exciting time where anyone can build from anywhere with internet access and AI assistance. He emphasizes curiosity, problem-solving, exploration, and learning to use AI as foundational skills.
- •Optimism grounded in practical limits of robotics and real-world automation
- •Personal context: growing up in East Germany shapes appreciation for change
- •AI helps fulfill the original “Sunday idea to evening app” dream of software
- •Skills for kids: curiosity, open-mindedness, independent problem-solving, AI fluency
- 19:04 – 21:48
If you fear AI taking your job: become the conductor + top AI tools to try
Thomas advises overcoming job anxiety by adopting AI deeply and becoming the expert who orchestrates agent workflows. He stresses responsible AI practices (testing, security, guardrails) and then shares his most-used tools: Copilot, ChatGPT, and transcription/summarization apps like Granola.
- •Best antidote to fear: upskill and become an AI power user
- •Humans remain orchestrators/conductors of multiple AI agents
- •Responsible AI: guardrails, red teaming, prompt-injection defense
- •Favorite tools: GitHub Copilot, ChatGPT, Granola (plus image generation for slides)
