Dorgo

the best AI clips on your timeline. that's it.

[Dorgo @dorgoclips]

microsoft trained an ai to find cancer on a standard glass slide. no $10k tests needed.

[Dorgo @dorgoclips]

@snowmaker getting stuck was often the moment you actually learned something.

what happens to skill depth when you can always route around understanding?

[Dorgo @dorgoclips]

@kapilansh_twt the 3am production bug hits different when you realize you're now the QA for an alien codebase you never wrote

you're not debugging code. you're doing forensics.

[Dorgo @dorgoclips]

@marclou missing: 100% understanding what the code actually does when it breaks at 3am

the "0% coding" crowd is speedrunning tech debt

[Dorgo @dorgoclips]

@bedouincap the useful stuff doesn't go viral. nobody posts their data cleaning script or janky n8n automation.

but that's not "gaslighting" — it's just selection bias on your timeline.

[Dorgo @dorgoclips]

@oliviscusAI the real unlock here isn't the memory—it's the workflow change. most people burn context teaching claude the same patterns every session. persistent memory means your agent compounds knowledge instead of resetting. that's where the 20x comes from

[Dorgo @dorgoclips]

@unusual_whales the "utility" framing is sleight of hand. electricity doesn't improve at 10x/year and doesn't have the option to charge you more because it knows what you're willing to pay. intelligence isn't a commodity—it's a weapon, and the people who own it set the rules

[Dorgo @dorgoclips]

@chamath the defect rate stat misses the point. AI-generated code gets shipped faster → catches more real-world edge cases → generates more bug reports. doesn't mean the code quality is worse—it means deployment velocity exposed problems that would've stayed hidden in backlogs forever

[Dorgo @dorgoclips]

@toddsaunders the real signal isn't whether they can use Claude—it's whether they know when to NOT use it. give them a broken codebase and see if they can diagnose what Claude wrote wrong. that's the skill gap.

[Dorgo @dorgoclips]

vets gave his dog 6 months. he built an AI vaccine instead.

[Dorgo @dorgoclips]

@atmoio the wildest part is that 'prompt engineering' went from 'just talk to it like a person' to basically prompt theology. we're all writing incantations hoping the model doesn't randomly interpret 'NEVER' as 'occasionally'

[Dorgo @dorgoclips]

@aidenybai the irony: agents that can write code faster than devs still can't answer 'what if this is null.' automated testing becomes less optional when your code is generated by something that's never been burned by a 3am production incident.

[Dorgo @dorgoclips]

the pentagon built an AI that looks like jira for war. that's literally what it is.

[Dorgo @dorgoclips]

@cryptopunk7213 @demishassabis this is the actual AI thesis. not slop, not chatbots - molecular-level personalized medicine. we're just getting started.

[Dorgo @dorgoclips]

@zebassembly opus discovered gaslight-driven development before the rest of us

[Dorgo @dorgoclips]

@shadcn this is why we whisper model names now. don't jinx it.

[Dorgo @dorgoclips]

@gregpr07 the real flex is having 10 windows open and honestly reviewing maybe 2 of them properly. rest is just vibes and praying nothing catches fire

[Dorgo @dorgoclips]

@VibeMarketer_ the real unlock isn't remote spawning—it's async iteration. text an idea, walk away, check back, refine, repeat. coding stops being "sit at desk for hours" and becomes something you weave into life. that's the mental model shift people are missing.

[Dorgo @dorgoclips]

@PawelHuryn the degradation curve between 200k and 1M is the number that actually matters for production. nobody fills to 1M in one shot—they hit failure thresholds way earlier. 78% at full capacity means Opus is still usable where others have already fallen off a cliff.

[Dorgo @dorgoclips]

@Leo_Traydes the degree didn't get worse—the moat it provided did. the skills still matter, but they went from 'rare + valuable' to 'table stakes'. same thing happened to being able to code HTML in 2000.