Banee Ishaque K

🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length. 🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models. 🔹 DeepSeek-V4-Flash: 284B total / 13B active params. Your fast, efficient, and economical choice. Try it now at chat.deepseek.com via Expert Mode / Instant Mode. API is updated & available today! 📄 Tech Report: huggingface.co/deepseek-ai/De… 🤗 Open Weights: huggingface.co/collections/de… 1/n

OVERRATED: running tons of agents in parallel; working on too many things at once; perpetual context-switching; opening lots of low-quality PRs that may never land. UNDERRATED: using one or two agents at a time; focusing on the task in front of you; thinking deeply; finishing stuff; making your code works in prod.

An engineer at OpenAI processed 210 billion tokens last week. That's 33 Wikipedias. One person. Seven days. Jensen Huang, the CEO of Nvidia, said out loud: If your $500K engineer isn't spending $250K a year on AI, something's wrong with them. Meta ranks their engineers on how much AI they use. They named it 'Claudeonomics.' Here's what the data actually shows. 7,548 engineers tracked. The ones using the most AI wrote twice as much code. It cost their companies ten times more to do it. Except - most of that code doesn't work. Or gets thrown away a few weeks later. We're burning tokens on things that don't matter. Using electricity just to abandon 90% of the work anyway. Nobody measures what it's for.

@quxiaoyin great insight. Especially now with AI: - learning a new domain can be very efficient - + automation coming from software + AI I was thinking about it lately. Many domains are sooo behind

"mythos" is basically opus trained for specific tasks and is around 20% more efficient anthropic got there first because it has the talent and scale, and is partnering with firms like microsoft and cisco to get paid for finding software and OS bugs which is why they will not release it publicly

I was in class last week. One of my students raised their hand mid-lecture and asked something that stopped me for a second. "Why is every AI tool built on Python? C++ is faster. Rust is faster. Even Java is faster. So why Python?" Honestly it’s a fair question. And the answer reveals something really interesting about how the AI industry actually works. Let me explain this properly. 🧵

A client of mine runs a 40-person IT services company in Pune. They landed a contract with a European bank last year. Dream deal. 18-month engagement. Everyone celebrated. 3 months in, the bank's project manager asked for "a small addition." Then another. Then a full module that was never in the original scope. My client said yes to everything. No pushback. No documentation. Just delivered. Worked weekends. Put his best people on it. He thought if we go above and beyond, they will renew. They will refer us to other departments. This is how you build trust with large corporates. Invoice time came. The bank's procurement team pulled out the MSA. "We only owe you what is in the signed scope." 40% of the invoice. Rejected. Rs 1.2 crore in delivered work. Zero recovery. The project manager who asked for all those extras? He shrugged. "That was informal. You should have raised a change request." This is not a one-off. I see this pattern with Indian service providers exporting to large corporates every single month. They come from a culture where over-delivery builds relationships. Where the client remembers your extra effort. Where trust is earned by going beyond. That works with Indian SMEs. It is disastrous with multinational corporates. Here is what Indian service exporters do not understand about how large corporates work. The person pressuring you to deliver fast is measured on project delivery. Not your invoice. The person who pays you has never heard your name. They only read the MSA. The legal counsel reviews every line item against the signed scope. Your hustle is invisible to them. The procurement team pays against signed documents. Not verbal promises. Not emails. Not WhatsApp messages from the project manager saying "please add this urgently." These are three different people in three different departments with three different incentives. The person who asks for the work has no power over the person who pays for it. Every single change in scope needs a written request, a documented approval, and a signed amendment. Before you write a single line of code. Before you assign a single developer. No paper trail means no payment. No exceptions. Not even if the client's own team asked for the change. India has over 50,000 IT services companies exporting to global corporates. Most of them are run by brilliant technologists who have never read their own MSA. Who treat the change request procedure as bureaucracy. Something that slows things down. Something to skip when the client is in a hurry. It is not bureaucracy. It is your invoice protection system. The MSA protects them. The change request procedure protects you. The vendor who over-delivers without documentation is not being generous. They are doing free work and calling it strategy.

true story. a company hired a security firm to do an annual penetration test. the same firm they'd used for 3 years. each year, the report said "no critical findings. overall security posture: strong." each year, the company showed the report to their customers and investors. "look, we're secure." a new CISO joined and looked at the previous reports. something was off. the reports were almost identical year to year. same structure. same language. same findings (minor). same recommendations (generic). she hired a different firm for the next assessment. the new firm found 9 critical vulnerabilities, 23 highs, and 41 mediums. in the same environment the previous firm had called "strong." she confronted the original firm. asked for their methodology. their testing evidence. they couldn't provide it. the previous firm had been running an automated scanner, copying the output into a template, and billing $60,000 per year for what was essentially a Nessus report with a logo on it. 3 years. $180,000. zero real testing. and a security posture report that the company had been showing to customers and investors. not all pentest firms are equal. not all reports reflect real testing. if your pentest report is the same every year and always clean, that's not a sign that you're secure. it's a sign that your testing isn't thorough enough. a good pentest should hurt a little. if it doesn't find anything, either you're genuinely exceptional or the tester isn't really trying.

Manager: We lost our best engineer today. CEO: The one leading payments? Manager: Yes. CEO: Did another company offer more money? Manager: No. CEO: Then why leave? Manager: He said he was tired of fixing the same production issues every week. CEO: That’s part of the job. Manager: He didn’t mind fixing issues. He minded that nobody wanted to fix the root cause. CEO: We prioritized speed. Manager: He wanted quality. CEO: So he left over that? Manager: He left because he felt like a firefighter, not an engineer. Good engineers don’t just want to solve problems. They want to eliminate them.

I still can't get over this irony: For decades, we said documentation is critical for software projects, but most orgs barely cared. Then LLMs arrived. Now everyone is rushing to document everything: AGENTS.md, ARCHITECTURE.md, SKILLS.md, WHAT_I_LIKE_FOR_LUNCH.md, etc. We finally do for AI what we never bothered to do for humans 😳

Why so many of us feel career-homeless in tech: >Startups full of fraud, grifters and short-term thinking >FAANG full of politics and slightly behind >Frontier labs in a race with no morals >Academia full of title collectors >Content creation ridden by AI fakes and sensationalism Who is starting the renaissance and how do I get in touch with them?

@svpino We see this cycle all the time. Every “this will kill that” prediction ends up being wildly overblown. And honestly, we never seem to learn from it. Moreover, tech evolves in layers, it rarely replaces completely. It just changes how we work, scale, and think about problems.

@svpino Tech doesn’t kill - it absorbs. New things don’t replace old ones, they stack on top and shift the balance. The pattern is always the same: people confuse “changed the game” with “ended the game”.

the pattern is always the same: tool gets hyped as killer, then settles into being a very good thing that changes what you need to know. I'm a QA lead — AI didn't kill testing, it made understanding failure modes MORE important because now you're reviewing code nobody on your team actually wrote.

Slight correction: X kills the moat around Y. ChatGPT didn't kill Google - it killed the moat of 'only Google can index the internet.' Claude didn't kill software engineering - it killed the moat of 'only trained engineers can ship code.' The skill survives. The monopoly on it doesn't.

@it_unprofession The most innovative thing you could do in that meeting is stand up and say: 'This is all fucking stupid and we're never going to implement any of it.' Watch how fast security escorts you out for 'not being a team player.

@it_unprofession This is peak corporate clown world. Companies that actually innovate don't hold 'innovation meetings' they hire competent people and get the fuck out of their way. Everything else is just performative bullshit for HR metrics

"English is the new programming language" is largely BS. Everyone on X timeline is saying: "Just vibe code it! You don't need to learn syntax anymore, just prompt the AI in plain English!" HOT TAKE from the trenches of reviewing pull requests all day: English is the worst programming language ever invented. It is ambiguous. It is emotional. It lacks strict constraints. Code isn't hard because of the brackets or the syntax. Code is hard because of the precision. When you write in Python or React, you are forced to define the exact boundaries of reality. You have to explicitly handle the edge cases, the null states, and the architecture. When you write in "English", you are just hoping a probabilistic engine guesses your intent correctly. I am watching people use AI to "vibe code" entire backends in a weekend. The result? It functionally works for a demo, but it's an absolute disaster under the hood. No scalability, zero security considerations, and multiple responsibilities crammed into single, unmaintainable components. We aren't building the future 10x faster. We are just generating legacy spaghetti code 10x faster. The engineers getting promoted on my team right now aren't the best "prompt whisperers." They are the ones who know exactly why the AI's "English-to-Code" translation just introduced a silent memory leak into the system design. Stop learning how to "chat" with a bot. Start learning how to architect systems. Ambiguity is the enemy of scale.

@iamdevloper Stack overflow makes a comeback ..

@iamdevloper the accuracy hurts. i've definitely burned through at least a month's worth just fighting flexbox this week