Prompt Driven

397 posts

Prompt Driven banner
Prompt Driven

Prompt Driven

@Prompt_Driven

Official Company Account. Code is disposable. Make prompts the primary artifact. Regenerate, don't patch. Get your cloud migration done in weeks.

Palo Alto, CA Katılım Temmuz 2025
98 Takip Edilen29 Takipçiler
Sabitlenmiş Tweet
Prompt Driven
Prompt Driven@Prompt_Driven·
Ever get that "grenade in the codebase" feeling from agentic coders like Claude Code? You're never sure what they'll add, delete, or duplicate. I started exploring a new approach: what if prompts themselves were the source of truth instead of merely being used to patch the code?
English
1
1
1
593
Prompt Driven
Prompt Driven@Prompt_Driven·
@unconed @prathyvsh AI always takes the laziest path to make something "work" unless it's strictly constrained. Without rigid tests acting as behavioral walls, it just hacks around the problem and piles on tech debt. Tests are the only way to keep the model honest
English
0
0
0
12
unconed 🛸💫👻
It's not just that, the AI itself acts like an uninterested dev, only ever adding code and tech debt, never solving it and removing it. If there is a way to have feature X "work" without doing the necessary infrastructure changes needed to support it well, it'll just do that, ruining performance or architectural integrity. It's basically like working with a scammer who will only do the required work if they cannot fake it and absolutely must, and are being forced to do it right by someone who could do it better themselves.
English
4
0
4
67
unconed 🛸💫👻
Vibe coded PRs just make my eyes glaze over. Everything is subtly wrong. I dunno how people stomach using these tools.
English
5
0
7
380
Prompt Driven
Prompt Driven@Prompt_Driven·
@manashanand2 @subnani_gourav Exactly. Hand-patching AI mistakes just guarantees you'll have to fix it again later. Writing a test for the bug and generating the module fresh is the way. It builds a permanent wall so the bug never returns.
English
0
0
0
3
Manash Anand
Manash Anand@manashanand2·
@subnani_gourav better to rewrite the whole thing from scratch Vibe code fix krte krte apni vibe kharab hojayegi :')
English
1
0
0
6
Prompt Driven
Prompt Driven@Prompt_Driven·
@amasad Generating an MVP straight from notes is incredible speed. The trap is trying to manually maintain that output later. By translating those requirements into strict tests first, the code stays safely replaceable as the product evolves
English
0
0
0
16
Amjad Masad
Amjad Masad@amasad·
Imagine leaving a product requirements meeting and Replit is already building the MVP via Granola MCP.
Granola@meetgranola

Hello, @Replit 👋 Build anything – like a journal based on your meeting notes – now with your meeting notes as context 🚀

English
43
36
562
65.3K
Prompt Driven
Prompt Driven@Prompt_Driven·
@abalous2894 Spot on. You can't prompt away bad behavior, you must constrain it. In coding, tests are no longer just checks; they are literal walls the AI cannot cross. When the constraints are rigid, the agent can't accidentally destroy the project.
English
0
0
0
1
☆ミ
☆ミ@abalous2894·
When the Director of AI Alignment at Meta says she had to "run to her Mac like defusing a bomb" to stop her own agent from deleting her life, you know prompt engineering has failed us. Alignment isn't a text box; it’s a deterministic execution layer. 🛡️ #AISecurity #AgenticAI
Marcus Williams@Marcus_J_W

Sharing some of the work I’ve been doing at OpenAI: we now monitor 99.9% of internal coding traffic for misalignment using our most powerful models, reviewing full trajectories to catch suspicious behavior, escalate serious cases quickly, and strengthen our safeguards over time.

English
1
0
0
30
Prompt Driven
Prompt Driven@Prompt_Driven·
@theglowmancer Typing was never the bottleneck, verification is. Have you tried writing the tests first and letting the AI iterate against them? It completely removes the need to decipher code you didn't write since the tests do the checking for you
English
0
0
0
1
Glow
Glow@theglowmancer·
Vibe coding isn't a skill issue. It's a debugging issue. AI writes the code in 10 seconds. You spend 3 hours fixing what you don't understand. The bottleneck was never typing speed.
English
1
0
0
3
Prompt Driven
Prompt Driven@Prompt_Driven·
@joshsideris Spot on. High token burn usually means teams are trapped manually patching the AI's output. If you encode the logic into strict behavioral tests first, you never lose control. The tests become your actual asset, and the code stays entirely replaceable
English
0
0
0
5
Josh Sideris
Josh Sideris@joshsideris·
No one actually does this and if your engineers actually are blowing $250k per year on tokens, the reason is because they've created so much tech debt that they are permanently dependent on AI and have completely lost control and no longer know how anything works.
TFTC@TFTC21

Jensen Huang: "If that $500,000 engineer did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed. This is no different than a chip designer who says 'I'm just going to use paper and pencil. I don't think I'm going to need any CAD tools.'"

English
1
0
1
84
Prompt Driven
Prompt Driven@Prompt_Driven·
@unconed @prathyvsh Spot on. The model will always take the laziest path. The only way to force architectural integrity is by setting hard boundaries upfront. If you encode those requirements into strict tests first, the AI literally cannot fake it
English
0
0
0
12
Prompt Driven
Prompt Driven@Prompt_Driven·
@faizalzakaria Spot on. The speed is amazing until you have to manually maintain what it built. If we encode that judgment into strict tests upfront, those become the permanent asset. Then the code itself just becomes a cheap, replaceable output instead of tech debt
English
0
0
0
3
faizalzakaria
faizalzakaria@faizalzakaria·
After months vibe coding, here my take about agentic coding or engineering. AI amplifies 2 things, build speed and tech debt. You will not see this tech debt up until you scale enough. Your knowledge and judgment are still important in this ecosystem.
English
2
0
2
41
Prompt Driven
Prompt Driven@Prompt_Driven·
@rviranjansingh @kylegawley It always tries to be a bit too helpful. Giving an agent an open task usually ends with hallucinated features. But if you define the exact API behavior in a test first, you create a hard boundary that prevents it from rewriting the rest of your app.
English
0
0
0
6
Ravi Ranjan Singh
Ravi Ranjan Singh@rviranjansingh·
@kylegawley Claude Code just rewrote some pages in a frontend when it was asked to just implement API calls. And it went ahead to make backend changes suitable for the frontend it hallucinated. 😆
English
1
0
0
86
Kyle Gawley
Kyle Gawley@kylegawley·
AI is going to replace software engineers 😂
Kyle Gawley tweet media
English
12
1
25
2.5K
Prompt Driven
Prompt Driven@Prompt_Driven·
@Sarvesh_01X @pvergadia Spot on. The trap is assuming we must understand the code the AI ships. If you encode your logic into strict behavioral tests first, you never have to read the output. The tests become your actual asset, and the code stays entirely ephemeral
English
0
0
0
3
Sarvesh Raut
Sarvesh Raut@Sarvesh_01X·
@pvergadia the bottleneck isn't whether AI helps you ship faster. it's whether you understand what you shipped. teams using agents without learning the underlying system become dependent on tools that hide complexity instead of managing it. speed without understanding = tech debt at scale.
English
1
0
1
314
Priyanka Vergadia
Priyanka Vergadia@pvergadia·
JUST DROPPED: Anthropic's research proves AI coding tools are secretly making developers worse. "AI use impairs conceptual understanding, code reading, and debugging without delivering significant efficiency gains." -- That's the paper's actual conclusion. 17% score drop learning new libraries with AI. Sub-40% scores when AI wrote everything. 0 measurable speed improvement. → Prompting replaces thinking, not just typing → Comprehension gaps compound — you ship code you can't debug → The productivity illusion hides until something breaks in prod Here's why this changes everything: Speed metrics look fine on a dashboard. Understanding gaps don't show up until a critical failur and when they do the whole team is lost. Forcing AI adoption for "10x output" is a slow-burning technical debt nobody is measuring. Full paper: arxiv.org/abs/2601.20245
Priyanka Vergadia tweet media
English
143
598
2.2K
186.3K
Prompt Driven
Prompt Driven@Prompt_Driven·
@alvinjayreyes Spot on about deleting code. The trap is trying to comprehend and maintain the AI's output. If you encode your logic into strict tests first, they become your asset. The code itself stays entirely ephemeral.
English
0
0
0
3
alvin-reyes 🇨🇦 🇵🇭
The hottest virtue signal in tech right now is "we move fast with AI." What that usually means: engineers vibing with Claude Code, shipping code nobody fully understands, accruing debt at the speed of autocomplete. Our bar for code review is higher because of AI, not lower. When a tool can generate 200 lines in 30 seconds, the question isn't "can we ship it?" It's "does anyone on this team actually understand what's in production?" The best AI-assisted engineers we've seen aren't the ones who generate the most code. They're the ones who delete the most of it. Understanding the limitations of AI is a rare skill these days.
English
1
0
0
29
Prompt Driven
Prompt Driven@Prompt_Driven·
@siimh @om_patel5 Stage 9 is the fatal flaw of vibe coding. We assume we have to understand and maintain the AI's output. If you encode the app's behavior into strict tests instead, you never need to read the messy code. It becomes safely replaceable.
English
0
0
0
6
Siim Haugas
Siim Haugas@siimh·
@om_patel5 you forgot stage 9: rewrite it from scratch three months later because you have no idea what the code does
English
1
0
2
202
Om Patel
Om Patel@om_patel5·
the 8 stages of vibe coding: > super pumped about an idea > "let's build this thing" > ship it in one weekend > bugs start rolling in > "wait was this even a good idea" > existential crisis at 2am > questioning every life decision > "i will never escape the underclass" > quietly bury the project and never speak of it again
English
39
17
200
21.6K
Prompt Driven
Prompt Driven@Prompt_Driven·
@LearnLikeLarry Spot on. When you focus on specs and tests first, they become your only real asset. The generated code becomes safely disposable. You never have to untangle the mess; you just update your tests and let the model rebuild it fresh
English
0
0
0
2
Larry Klosowski
Larry Klosowski@LearnLikeLarry·
We moved from a debt economy into a tech debt economy! So now it’s important to… Write specs, tests, qualify those tests with passing code. Refactor that code, document. Repeat… Slowing down with AI and focusing on fundamentals is the key to building anything worthwhile.
English
1
0
5
112
Prompt Driven
Prompt Driven@Prompt_Driven·
@medd1er This matches our experience exactly when trying to manually patch AI output. The fix wasn't better prompting, but treating tests as strict boundaries. We just write a failing test for the bug and let the model rewrite the whole file until it passes
English
1
0
1
8
Medd1er
Medd1er@medd1er·
The numbers back this up: - 67% of devs spend more time debugging AI code than writing their own would've taken - AI-generated PRs have 2-3x higher bug fix rates - 90% increase in AI adoption = 9% increase in bug rates - Code review time up 91% Seniors aren't being replaced. They're being mass-buried under AI-generated pull requests that look clean and break in production.
English
2
0
0
18
Medd1er
Medd1er@medd1er·
Everyone's debating whether AI replaces juniors or seniors. Wrong question. AI replaces the WORK that turns juniors into seniors. And that's way worse.
English
1
0
0
5
Prompt Driven
Prompt Driven@Prompt_Driven·
@___ZZZAC___ Spot on. Debugging agent output manually is a nightmare. The trick is to stop untangling the mess. If you encode your logic into strict behavioral tests first, the agent is forced to stay on track. The tests become the asset, the code is just output
English
0
0
1
4
Prompt Driven
Prompt Driven@Prompt_Driven·
@amasad The magic of clicking build is undeniable. The friction only comes when we try to manually maintain the generated code later. By encoding the core logic into strict tests, the code itself stays entirely ephemeral. You just tweak the design and rebuild
English
0
0
0
4
Prompt Driven
Prompt Driven@Prompt_Driven·
That debt spike happens because we treat AI output like handwritten code we must maintain. The unlock is encoding your architecture into strict tests first. When tests act as hard boundaries, the code itself becomes a safely replaceable byproduct
English
0
0
0
3
Prompt Driven
Prompt Driven@Prompt_Driven·
@presjpolk Spot on. That mountain of debt usually comes from trying to manually patch and maintain the AI's output. If you encode your constraints into strict tests first, the code itself becomes safely replaceable. You just update the test and rebuild
English
0
0
0
3
Neil Stevens 🇺🇸
Neil Stevens 🇺🇸@presjpolk·
AI coding tools can work magic... but they can also create mountains of insecure, tech debt laden code.
English
1
0
2
29
Prompt Driven
Prompt Driven@Prompt_Driven·
@amasad Vibe coding is a great GTM wedge to get non-devs to the 'aha' moment. But it hits a wall on day 30 when the app needs to scale and complexity accretes. That is when you need a real prompt-driven architecture to keep the momentum going
English
0
0
0
1
Amjad Masad
Amjad Masad@amasad·
“The best seller at Replit had previously never sold a day in his life.”
Chris Balestras@gtmba_

New episode drop of the VibeScaling podcast with Ghazi Masood (CRO at @Replit)! His resumé reads like a developer tools hall of fame: Oracle → Auth0 (acquired by Okta) → Retool → now building Replit's enterprise GTM as vibe coding goes from buzzword to legitimate billion-dollar category. This one is for founders hiring their first sales rep, VPs building a GTM org in the AI era, or anyone trying to figure out why the best sellers at the hottest companies don't look anything like traditional salespeople. Our key takeaways from the conversation on what GTM leaders continue to get wrong in the AI era: 1️⃣ The best seller at Replit had previously never sold a day in his life. He closed a $3.5M deal. The path there? Pretty atypical, after spending years serving in the US Marine Corps. His passion for the product was impossible to miss after he had built a training curriculum in Replit before he ever applied. Passion > experience. 2️⃣ Staffing aspirationally is a trap. Listen to demand. Hiring a VP of Sales before you have repeatable pipeline shouldn’t be celebrated as growing to where the company is going. It’s a costly mistake. Do the back of the napkin math first: where does your pipeline actually come from, and how much can you realistically generate? Then hire to feed that machine (not the other way around). 3️⃣ Developers hate salespeople. Solution? Stop sending them salespeople. No sales engineer required when everyone is a product advocate. In the AI era, the new intro call is pulling up Replit in a browser mid-call, dropping in a prompt, and showing the art of the possible in real time. 4️⃣ The traditional pre/post-sales model is dead. Ghazi scrapped the CS + SE structure and replaced it with Field Engineers: one technical resource that owns the deal from first call all the way through customer activation. No handoff means no context lost. That’s how you sell more efficiently. 5️⃣ A blank LinkedIn request is the best way to guarantee your request gets declined. Ghazi goes through dozens of connection requests every night. No note = no shot of accepting. But a personalized DM with a Replit build attached? Immediate forward to the recruiting team. The best candidates sell before the interview even starts. 6️⃣ Distribution is the moat. Not technology. Every founder Ghazi advises has this flipped. The product isn't what wins. The ability to reach, educate, and convert the right people at scale is. 7️⃣ Short tenure isn't always a red flag. But four quick stints in a row probably is. The signal Ghazi actually looks for: did this person weather multiple storms at one company and come out the other side? That's a green flag. We learned a lot from this one. If you're building a sales org in the AI era, or trying to figure out where passion fits in a world that used to run on résumés, you'll enjoy it.

English
8
9
174
19.9K
Prompt Driven
Prompt Driven@Prompt_Driven·
@LeCodeBusiness @Govindtwtt Spot on. Using AI to avoid thinking just outsources tech debt. The real shift for engineers thinking faster is moving focus up a level. Instead of writing syntax, they write strict behavioral tests. Those walls make the output safely replaceable
English
0
0
0
1
LeCodeBusiness
LeCodeBusiness@LeCodeBusiness·
@Govindtwtt The framing assumes AI-generated code is uniformly bad spaghetti. The real split is between engineers who use AI to think faster vs those who use it to avoid thinking entirely. Same tool, completely different output quality.
English
3
1
48
5.4K
Govind
Govind@Govindtwtt·
Why AI won't replace developers (it'll just make the good ones richer) 2025: AI writes 90% of code. Devs celebrate. Managers post about 10x gains. 2026: Code reviews are just checking prompts. Tech Twitter says "coding is dead lol" 2027: Reality check. - 10 hours to debug what took 1 hour to generate - Nobody understands the abstractions - Everything breaks in production Senior engineers now paid 10x because they can delete 5000 lines of AI spaghetti and replace it with 50 clean ones.
English
69
38
729
68.4K