Software engineering is dead now - AI & Theo from t3dotgg

SUMMARY

A developer-creator argues AI has made code nearly free, forcing smaller teams and layoffs; he demos building “Lawn” fast, analyzes Jack Dorsey’s Block cuts, urges developers adapt.

IDEAS:

• The “six months from AI taking developer jobs” meme feels suddenly real, not hypothetical anymore.
• Many industry signals suggest software creation economics changed, from coding scarcity to abundance rapidly.
• Cloudflare reportedly rewrote Next.js JavaScript in a week using AI, reshaping expectations dramatically.
• CEOs who barely code can now build full apps, cloning billion-dollar companies quickly today.
• Jack Dorsey cut Block nearly in half despite profitability, implying structural technological pressure on labor.
• Media creators may feel safer, yet daily engineer messages indicate widespread fear and dislocation.
• AI code review becomes essential because vibe coding generates huge PRs humans won’t thoroughly inspect.
• Tools with memory that learn feedback reduce review noise, focusing attention on signal over chatter.
• AI reviewers catch subtle repository hygiene issues like missing gitignore entries within massive changes.
• Copy-pastable prompts turn review findings into immediate fixes, compressing feedback loops substantially.
• Multi-language repos spanning Rust, Swift, TypeScript, Electron become manageable with AI comprehension.
• Async cleanup and promise misuse can be caught by AI reviewers better than alternatives sometimes.
• Building an internal replacement becomes easier than persuading vendors to fix long-standing issues.
• Lawn replicates much of Frame.io functionality in two weeks, part-time, without writing code.
• Product velocity now depends more on taste, specification, and constraints understanding than typing speed.
• Open-sourcing a product becomes less risky because cloning is inevitable when implementation costs collapse.
• Ideas become simultaneously more and less valuable when execution is faster and copying easier.
• Adobe bought Frame.io for $1.3B, yet comparable utility can be rebuilt quickly by individuals now.
• Great UX can be achieved with AI-generated UI and assets if interaction performance is prioritized.
• Pre-warming and hover subscriptions show humans still supply winning patterns, then agents propagate consistently.
• The worst these tools will ever be is today; improvements will continue making building easier.
• Code can be rewritten in days, but partnerships, legal agreements, and data moats remain durable.
• Financial and music streaming businesses rely on negotiations and infrastructure not replicable by vibe coding.
• Smaller, flatter teams can outperform larger organizations because coordination overhead becomes dominant constraint.
• Mythical Man-Month becomes more severe when PR volume explodes and review becomes bottleneck.
• Engineering bottleneck shifts from writing code to reviewing, testing, QA, releasing, and rollback safety.
• Lines of code become free; tests and verification become the scarce resource for shipping reliably.
• Workflow changes like “who owns main today” enable rapid shipping with minimal people.
• Traditional funnel from problem to shipped solution collapses when screenshots can become fixes via agents.
• Organizations will ask why everything gets implemented but nothing ships, revealing process misalignment.
• Many elite engineers are terrible at QA because they don’t use products like real users.
• Big teams are aircraft carriers: powerful throughput, but slow to change direction under new conditions.
• Massive companies struggle shipping trivial model additions because bureaucracy outweighs technical difficulty.
• Companies not forcing adoption of AI dev tools are “already doomed,” creating opportunities for competitors.
• Developers must shift toward user understanding, release involvement, automation, and ownership for survival.
• Agents lack initiative and agency; humans who provide direction and context remain valuable longer.
• If an agent knows more about your customers than you do, your role becomes redundant fast.

INSIGHTS

• When code generation becomes cheap, organizational coordination and verification become the true competitive bottlenecks.
• The enduring moat shifts from implementation skill toward distribution, partnerships, data rights, and regulated infrastructure.
• AI collapses the idea-to-product cycle, making novelty powerful while simultaneously making copying inevitable.
• Layoffs in profitable firms signal redesigning for speed, not merely cutting costs during downturns.
• Review, QA, and release engineering become core craftsmanship as typing code becomes commoditized labor.
• Small teams win by reducing approvals, minimizing ownership boundaries, and iterating tightly around real users.
• Managers who can specify well become force multipliers because specification now directly produces working systems.
• Bureaucracy makes simple changes “hard,” creating openings where lean competitors can out-ship giants rapidly.
• Human advantage becomes agency: noticing problems, choosing direction, and caring about outcomes beyond prompts.
• Customer empathy becomes job security; ignorance of users lets models replace developers who only implement tickets.
• Open source becomes less threatening when differentiation moves to experience, speed, and relationships rather than code.
• The new developer identity resembles product operator: shipping, testing, rollback, and reliability over clever abstractions.

QUOTES:

• “We’re 6 months away from AI taking our jobs as developers forever.” — Speaker
• “I’m starting to really feel it.” — Speaker
• “This is terrifying. We are there now.” — Speaker
• “I’m vibe coding a bit close to the sun lately.” — Speaker
• “Code Rabbit has reviewed pretty much every line of code I’ve written in the past few months” — Speaker
• “I’ve tried all of these AI code review tools with my team and this is the one my team chooses every single time.” — Speaker
• “Being that this PR was like 3,000 lines of code, there’s no world in which I would have bothered to read it” — Speaker
• “I was able to build it in 2 weeks.” — Speaker
• “I have not written a single line of code in this project.” — Speaker
• “Ideas haven’t been valuable for a long time.” — Speaker
• “Everything has changed.” — Speaker
• “This is the worst it’s ever going to be, guys.” — Speaker
• “It’s not worth trying to convince somebody when I can just out build them now.” — Speaker
• “Today, we’re making one of the hardest decisions in the history of our company.” — Jack
• “We’re reducing our organization by nearly half from over 10,000 people to just under 6,000.” — Jack
• “We’re not making this decision because we’re in trouble. Our business is strong.” — Jack
• “Repeated rounds of cuts are destructive to morale to focus and to the trust” — Jack
• “I’d rather it feel awkward and human than efficient and cold.” — Jack
• “We’re going to build this company with intelligence at the core of everything we do” — Jack
• “Lines of code effectively are free now.” — Speaker
• “Tests matter. Tests matter a lot.” — Speaker
• “Now you can literally take the user problem, paste the screenshot into the agent, and skip all the way down to here.” — Speaker
• “Most engineers are really bad at this.” — Speaker
• “Being paid to type in lines of code is over.” — Speaker
• “If the agent knows more about your customers than you do, you don’t have a job anymore.” — Speaker

HABITS

• Use AI code review on every pull request to catch mistakes hidden inside massive diffs.
• Train review tools with feedback so they remember preferences and stop repeating irrelevant comments.
• Copy-paste generated prompts from review feedback to implement fixes quickly and ship confidently.
• Open-source projects occasionally to validate code quality and invite external scrutiny and improvement.
• Build internal alternatives when vendors ignore critical issues rather than spending months persuading them.
• Work on products in background time, leveraging agents to maintain progress without full-time focus.
• Establish reusable performance patterns, document them, then let agents propagate changes across codebase.
• Maintain multiple worktrees and threads for experiments, accepting many will remain untested initially.
• Rotate “main branch ownership” daily so one person ships continuously while others develop features.
• Prioritize real-time UX responsiveness by pre-warming routes and subscribing on hover for perceived speed.
• Use the product daily to notice friction, then immediately spec fixes to an agent pipeline.
• Keep communication channels open during transitions so teams can say goodbye and share context.
• Provide generous severance buffers to reduce stress and allow thoughtful job searches after layoffs.
• Audit release processes, rollback systems, and alerting regularly because shipping speed increases breakage risk.
• Spend more time talking to users, designers, and product teams than obsessing over code elegance.
• Automate your own tasks intentionally to discover adjacent responsibilities that remain distinctly human work.
• Evaluate companies by whether they force adoption of modern AI tools, signaling adaptability and survival.
• Avoid over-hiring for “more code” output; hire for testing, QA, release engineering, and ownership.
• Treat review and QA as primary craft skills, practicing them deliberately rather than avoiding them.
• Keep hardware access resilient; ensure personal computing isn’t solely dependent on employer-provided devices.

FACTS:

• Block reduced headcount from over 10,000 to just under 6,000 people overall.
• Block’s reduction means over 4,000 people asked to leave or enter consultation process.
• Block offered affected employees 20 weeks salary plus one week per year of tenure.
• Block offered equity vested through end of May alongside six months of healthcare coverage.
• Block offered corporate devices and $5,000 to support transition needs after layoffs.
• Frame.io was acquired by Adobe in 2021 for $1.3 billion dollars.
• Block operates Square, Cash App, Afterpay, Tidal, BitKey, and Proto businesses.
• Square and Cash App generate substantial revenue that funds most of Block’s operations.
• Financial products rely on agreements and partnerships that cannot be recreated quickly by coding.
• Streaming music catalogs depend on negotiations, payment structures, and rights management beyond software alone.
• Tidal provides unusually in-depth track credits compared with other streaming platforms, according to speaker.
• “Mythical Man-Month” describes adding people can slow projects due to coordination overhead.
• Large organizations can struggle to ship even one-line model additions because approval processes dominate.
• Copilot in VS Code included Gemini 3.1 Pro earlier than Copilot CLI, per discussion.
• Speaker’s Lawn project earned roughly $350 MRR and might lose money currently.
• Speaker employs about 15 people across projects while running a channel and company concurrently.
• Lawn codebase is about 16k lines, mostly TypeScript, created largely by AI agents.
• T3 Chat is nearing 100,000 lines of code, showing scale achievable quickly now.
• T3 Code reached 52,581 lines within two to three weeks, built via vibe coding.
• Writing code shifted from most expensive to cheapest step, making testing and release dominant costs.

REFERENCES

• Meme: “6 months away from AI taking our jobs as developers forever”
• Baby Keem
• OpenClaw
• “top G”
• Cloudflare
• Next.js
• Jack Dorsey (original CEO of Twitter)
• Block
• Square
• Cash App
• Afterpay
• Tidal
• BitKey
• Proto
• CodeRabbit (AI code review tool)
• gitignore / build directory
• Rust
• Swift
• TypeScript
• Electron
• Frame.io
• Adobe
• Premiere Pro
• Final Cut
• T3 Code
• T3 Chat
• Codex app
• OpenAI
• Copilot
• VS Code
• Gemini 3.1 Pro
• Gemini CLI
• Copilot CLI
• Burke (VS Code/Copilot at Microsoft)
• Microsoft
• Google
• Twitch
• Cobra (COBRA healthcare)
• Linear
• Notion
• Discord
• “Mythical Man-Month”
• “launch week”
• “agent MD”
• “whisper flowing”
• “pre-warming and subscriptions on hover”

ONE-SENTENCE TAKEAWAY

AI makes code cheap; developers must own users, QA, and shipping to stay relevant.

RECOMMENDATIONS

• Adopt AI code review immediately, because massive diffs hide simple mistakes humans won’t catch.
• Give tools feedback repeatedly so they learn your standards and reduce noisy, repetitive suggestions.
• Treat tests as your moat; invest in integration coverage and reliability harnesses relentlessly now.
• Rebuild internal tools when vendors ignore needs; shipping yourself is often faster today.
• Minimize team size and approvals; optimize for tight ownership and fast iteration cycles.
• Document performance patterns clearly, then instruct agents to apply them across the entire codebase.
• Measure success by shipped outcomes, not lines of code, commits, or PR counts.
• Shift developer time toward QA, release planning, rollback automation, and alerting for safety.
• Start talking to real customers weekly; learn pain points models can’t infer from tickets alone.
• Use your own product daily to detect friction, then spec improvements directly to agents.
• Automate your current tasks proactively to discover higher-leverage work surrounding the automation boundaries.
• Evaluate employers by AI tool adoption; laggards will likely downsize when they finally react.
• Reduce bureaucracy around model updates; one-line integrations should not require multi-team approval chains.
• Design severance and communication processes humanely; morale and trust matter during organizational shocks.
• Build defensible moats beyond code: partnerships, data rights, compliance, and distribution relationships endure.
• Open-source selectively to validate quality, build trust, and accelerate adoption in competitive markets.
• Prioritize UX speed with prefetching and hover subscriptions; delight becomes differentiation when cloning is easy.
• Rotate ownership of mainline shipping so someone is always integrating, testing, and releasing daily.
• Stop fetishizing craftsmanship of typing; embrace craftsmanship of specification, verification, and operations instead.
• If your product is only code, add business constraints and relationships that competitors can’t replicate quickly.

---

📇 Additional Metadata

• 🗂 Type::note
• 🏷️ Tags::
• 📡 Status::🌱