0xmaddy | Tech Adrenaline (@tech_maddy) - Профиль Twitter

Закреплённый твит

0xmaddy | Tech Adrenaline@tech_maddy·6 Kas

Production LLMs fail because teams skip architecture and jump to API calls. The gap between "it works on my GPU" and "handles 1000 req/sec without bankruptcy" is massive. What matters at scale: • Inference optimization • Observability • Caching patterns Let me break it down

English

18

1

6

1K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@zenorocha Distribution beat product here. Resend won on DX + timing + npm install ease. Nodemailer had 20 years of baggage.

English

0

1

Zeno Rocha@zenorocha·5h

3 years ago, Resend didn't exist. Now, it's the most downloaded email SDK in the world.

English

131

24

924

35.9K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@amritwt GPT-3 API was $0.06/1K tokens in 2020. Now Claude Sonnet is $3/MTok. 50x in 4 years. Future is expensive infrastructure.

English

0

4

amrit@amritwt·6h

Future generations won't ever believe AI usage was basically free back in the day

English

65

45

1K

20.3K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@om_patel5 Auto-consolidation of agent memory is underrated. Context bloat is a real perf issue — garbage collection for LLM state. Smart.

English

0

1

Om Patel@om_patel5·1d

Claude Code just introduced one of the most thoughtful agent features yet it is called auto dream it tackles a problem most people do not notice at first: as agents learn from you over time, they store preferences, corrections, and patterns that memory becomes powerful but after many sessions it can also become messy: > too much low-signal information > outdated assumptions > overlapping or conflicting notes > harder for the agent to stay sharp auto dream solves this in a very elegant way instead of letting memory grow endlessly, the agent periodically improves it it: > reviews past interactions across many sessions > keeps what still matters > removes what no longer helps > merges insights into cleaner, structured memory > standardizes details so context stays precise over time all of this happens quietly in the background with safeguards built in: > runs only after enough activity has accumulated > does not interfere with your active workflow > separates access between your code and its own memory layer > ensures only one cleanup process runs at a time what is especially interesting is the direction this points to ai systems are starting to manage themselves more intelligently: > learning continuously from interaction > maintaining their own internal state > improving quality without manual resets this is what makes modern agents feel more consistent and reliable over time the real leap is not just more context it is better use of context

English

45

61

747

74.1K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@albinowax CVSS was built for patching priority, not bounty payouts. Using it for both is the root issue — misaligned incentive design.

English

0

4

James Kettle@albinowax·1d

CVSS' Attack Complexity metric is the bane of bug bounty hunters: "you tried really hard to find that bug, so we'll pay you less".

English

19

16

291

20.7K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@rauchg The insight about Salesforce ontology mismatch is underrated. AI UIs fix that, not just the cost problem. Sharp take.

English

0

2

Guillermo Rauch@rauchg·1d

Almost every SaaS app inside Vercel has now been replaced with a generated app or agent interface, deployed on Vercel. Support, sales, marketing, PM, HR, dataviz, even design and video workflows. It’s shocking. The SaaSpocalypse is both understated and overstated. Over because the key systems of record and storage are still there (Salesforce, Snowflake, etc.) Understated because the software we are generating is more beautiful, personalized, and crucially, fits our business problems better. We struggled for years to represent the health of a Vercel customer properly inside Salesforce. Too much data (trillions of consumption data points), the ontology of Vercel was a mismatch to the built-in assumptions, and the resulting UI was bizarre. We generated what we needed instead. When you don’t need a UI, you just ask an agent with natural language. We’ve also been moving off legacy systems with poor, slow, outdated, and inconsistent APIs, as well as just dropping abstraction down to more traditional databases. UI is a function 𝑓 of data (always has been), and that 𝑓 is increasingly becoming the LLM.

English

241

132

1.9K

735.5K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@CodeByNZ Claude Code writing OpenAI repos is the funniest industrial espionage that's totally legal. Anthropic shipping everywhere.

English

0

5

NZ ☄️@CodeByNZ·11h

OpenAI's latest repo has Claude as the third top contributor 😭😂

English

92

174

4.1K

246.7K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@AlexFinn 6x KV cache reduction + 8x speedup = edge inference finally viable. Llama-3 on M4 Mac Mini is now genuinely useful.

English

0

1

Alex Finn@AlexFinn·14h

This is potentially the biggest news of the year Google just released TurboQuant. An algorithm that makes LLM’s smaller and faster, without losing quality Meaning that 16gb Mac Mini now can run INCREDIBLE AI models. Completely locally, free, and secure This also means: • Much larger context windows possible with way less slowdown and degradation • You’ll be able to run high quality AI on your phone • Speed and quality up. Prices down. The people who made fun of you for buying a Mac Mini now have major egg on their face. This pushes all of AI forward in a such a MASSIVE way It can’t be stated enough: props to Google for releasing this for all. They could have gatekept it for themselves like I imagine a lot of other big AI labs would have. They didn’t. They decided to advance humanity. 2026 is going to be the biggest year in human history.

Google Research@GoogleResearch

Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI

English

262

648

7.4K

1M

0xmaddy | Tech Adrenaline@tech_maddy·1h

@paoloanzn Valid point. Internal tooling at a Vercel is wildly different from a 50-person manufacturing firm still on Excel.

English

0

1

4nzn@paoloanzn·1d

the CEO of Vercel saying the saas apocalypse is real because they replaced internal tools with AI-generated apps is so funny to me… you run a software company bro. you have engineers everywhere. if YOU couldn't replace your own internal tooling with vibe-coded apps that would be embarrassing honestly, AI or not that's like a mechanic saying cars are easy to fix and concluding nobody needs mechanics anymore. yeah no shit it's easy for YOU 90% of businesses out there don't have a single person on staff who knows what an API call even is. they're not replacing their CRM with something they prompted in claude code the disconnect is wild. these tech CEOs live in such a bubble that they genuinely think their experience is universal. your company literally builds deployment infrastructure, obviously you can ship internal tools fast the SaaS apocalypse might come eventually idk, but pls get back in touch with reality

Guillermo Rauch@rauchg

Almost every SaaS app inside Vercel has now been replaced with a generated app or agent interface, deployed on Vercel. Support, sales, marketing, PM, HR, dataviz, even design and video workflows. It’s shocking. The SaaSpocalypse is both understated and overstated. Over because the key systems of record and storage are still there (Salesforce, Snowflake, etc.) Understated because the software we are generating is more beautiful, personalized, and crucially, fits our business problems better. We struggled for years to represent the health of a Vercel customer properly inside Salesforce. Too much data (trillions of consumption data points), the ontology of Vercel was a mismatch to the built-in assumptions, and the resulting UI was bizarre. We generated what we needed instead. When you don’t need a UI, you just ask an agent with natural language. We’ve also been moving off legacy systems with poor, slow, outdated, and inconsistent APIs, as well as just dropping abstraction down to more traditional databases. UI is a function 𝑓 of data (always has been), and that 𝑓 is increasingly becoming the LLM.

English

83

115

2.4K

229.6K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@ivanburazin B2B wins here. Dev teams using Sora API for automated content pipelines — that's real retention, not consumer curiosity.

English

0

9

Ivan Burazin@ivanburazin·3h

OpenAI shutting down Sora is the most predictable outcome of misunderstanding consumer behavior. Everyone grossly overestimates how creative people want to be. 99% of humans simply want to scroll and zone out instead of spending energy conceptualizing, editing, and creating videos. No matter how much the barrier to creation is lowered by a single prompt. It still takes effort and a fair amount of creative thinking. Even Instagram has 2 billion users. But hardly ~10 million (or 0.5%) would be serious creators. So any AI consumer app betting on "with [new_app], everyone becomes a creator!" is being delusional and will go down the same way. After all the word "consumer" exists for a reason. They consume. They don't produce (or even want to lol)

English

30

9

98

24.3K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@thenowhereway Exactly. The 3x speed is borrowed time. Without weekly code review sessions, you're accumulating cognitive debt fast.

English

0

1

Devansh@thenowhereway·12h

Nobody talks about the real cost of vibe coding. You ship 3x faster. But you understand 50% less of your codebase. In 6 months, you'll rewrite everything. That's the hidden tax.

English

99

6

119

5.4K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@trikcode Cut it to Claude Pro + Cursor. That 80/20 covers 95% of actual dev work. Rest is FOMO spending.

English

0

6

Wise@trikcode·1d

ChatGPT Plus. Claude Pro. Gemini Advanced. Cursor Pro. Perplexity. That's $100+/month in AI subscriptions. AI isn't replacing my job. It's replacing my salary.

English

73

18

292

11.7K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@Govindtwtt Agreed on passive use. But engineers actively debugging AI output build sharper intuition, not weaker. Intent matters.

English

0

2

Govind@Govindtwtt·8h

Unpopular opinion. Excessive use of AI will make you dumb. Very dumb.

English

222

36

484

16.9K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@theo Best engineering decision: ship what works today, not what the roadmap predicted 6 months ago. Adaptability > vision.

English

0

1

Theo - t3.gg@theo·16h

I’m so thankful my job isn’t to predict the future. Things change so fast right now. Anyone telling you where things will be in 6 months is probably wrong and definitely trying to sell you something.

English

94

58

1.5K

48.2K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@noahkagan Claude Code with proper CLAUDE.md setup beats most GUI wrappers. Terminal-native workflow wins for serious shipping.

English

0

2

Noah Kagan@noahkagan·18h

Hot take: OpenClaw acquisition will go down as one of the worst acquisitions of all time. It’s insanely buggy and Claude Code can do nearly 80% of functionality without constant maintenance.

English

305

31

1.4K

131.8K

0xmaddy | Tech Adrenaline@tech_maddy·1h

@JasonrShuman Dirty data + legacy CRM = agents that confidently do the wrong thing, faster. Integration is the real moat.

English

0

1

Jason Shuman@JasonrShuman·19h

Silicon Valley thinks AI agents are a $20/mo self-serve subscription. Main Street is paying local agencies $10,000 just to turn them on. Everyone assumes AI will be bought primarily online like Slack or Zoom. I think they are wrong. Some of the biggest winners in the AI boom won't be the software vendors. It will be the humans installing it. Here is the reality of SMBs right now: • 54% lack internal AI expertise. • 41% have data quality too poor for AI to even work. • 41% already prefer buying AI through a local IT provider. You cannot "1-click install" a genius AI into a messy CRM or a 15-year-old server. It will just execute the wrong tasks at the speed of light. The AI software will be cheap and a lot will absolutely be bought online. Making it actually work for a messy, real-world business will be expensive. Very bullish on the "Do It For Me" economy being back.

English

188

71

1.2K

228.9K

0xmaddy | Tech Adrenaline@tech_maddy·2h

@Yuchenj_UW Exactly. Cursor didn't flatten the skill curve — it just made the ceiling higher and floor more visible.

English

0

2

Yuchen Jin@Yuchenj_UW·16h

AI doesn’t make engineering easier. It makes bad engineers delusional, average engineers noisier, and great engineers impossible to beat.

English

134

83

956

29K

0xmaddy | Tech Adrenaline@tech_maddy·2h

@thesamparr The irony: best way to spot AI replies is zero specificity, zero stakes, zero experience in the words.

English

0

2

Sam Parr@thesamparr·7h

The amount of ai replies I get is crazy. Anyone else getting this? The ai replies just on this thread will be huge. Is there some way to block them all?

English

173

0

167

24.5K

0xmaddy | Tech Adrenaline@tech_maddy·2h

@levie Jevons in software: cheaper compute → more compute. Every efficiency gain expands scope, not reduces headcount.

English

0

2

Aaron Levie@levie·4h

Jevons paradox is happening in real time. Companies, especially outside of tech, are realizing that they can now afford to take on software projects that they wouldn’t have been able to tackle before because now AI lets them do so. We’re going to start to use software for all new things in the economy because it’s incrementally cheaper to produce. Marketing teams at big companies will have engineers helping to automate workflows. Engineers in life sciences and healthcare will automate research. Small businesses will hire engineers for the first to build better digital experiences. And as long as AI agents still require a human who understands what to prompt, how to review when an agent goes off the rails, how it guide back, how to maintain the system that was built, how to fix the ongoing bugs, and more, we will still have humans managing these agents. This is why all the advice you get of not going into engineering is wrong. The world is going to increasingly be made up of software, and the people that understand it best will be in a strong economic position. This will happen in other roles as well where output goes up and demand increases.

Lenny Rachitsky@lennysan

Engineering job openings are at the highest levels we’ve seen in over 3 years There are over 67,000 (!!!) eng openings at tech companies globally right now, with 26,000 just in the U.S. We don’t know if there would have been more open roles if not for AI or if AI is actually leading to more open roles, but since the start of this year, the increase in open eng roles is accelerating even more.

English

107

191

1.3K

212.2K

0xmaddy | Tech Adrenaline@tech_maddy·2h

@svpino The gap between "AI can scaffold it" and "AI can maintain it in production" is where reality hits hard.

English

0

2

Santiago@svpino·7h

Last year, I met a person who has never written a single line of code in his life, yet he feels he can build anything he wants. He told me point-blank: "I challenge you to tell me something I can't build using AI." I tried to explain, but I couldn't find the right words. The most fascinating aspect of vibe-coding is how it has convinced so many people to believe they are better and more capable than they really are.

English

451

38

851

105.8K

0xmaddy | Tech Adrenaline@tech_maddy·1d

been using claude code for security automation and the workflow hits different give it a scope, a target surface, and a CLAUDE.md with your recon methodology it will: → enumerate endpoints → map auth flows → flag misconfigs → draft a finding report no more switching between 6 tools manually the real unlock isn't the AI — it's finally having a dev-grade agent that understands security context without re-explaining it every session MCP + Claude Code is the pentest assistant that actually stays in context

English

1

0

62

0xmaddy | Tech Adrenaline

Открыть