Archive, page 4 of 9
Articles 151–200, newest first.
When Not to Use AI: A Sarcastic Poem Worth Reading
A prose poem tells you to use AI for everything, and means the opposite. Here is the line I use to decide what to hand a model and what to keep.
86,000 exposed BMCs: your server's hidden second computer
New Black Hat research found critical bugs in baseboard management controllers from HPE, Dell, Supermicro and more. Why this matters most if you buy used enterprise servers.
Google Assistant shuts down Sept 4: what it means for you
Google Assistant is being removed from Android phones, tablets and paired devices from September 4th. Here's what actually breaks, and why an email is all the notice a platform owes you.
Ted Lasso Season 4 and the Case for One Anchor Product
Apple TV's 2026 slate is a portfolio bet: one reliable anchor plus several risky swings. That structure is worth copying if you ship software alone or run a small team.
EA's $55bn Sale: What a Record Buyout Means for Devs
The Electronic Arts sale closed at $55bn, funded partly with $20bn of borrowed money. Here's what that debt does to a company's roadmap, and why it matters to small teams.
Korea capped 2x leverage. Retail bought 3x instead.
Korea tightened rules on 2x single-stock products and $1B more flowed into US 3x ETFs. It is a systems design failure, and engineers ship the same bug.
A SIMD Viterbi decoder in Rust: reading the benchmarks properly
A SIMD Viterbi decoder in Rust beats libfec on every code in its README. But the honest speedup depends entirely on which libfec build you compare against.
FIPS 140-3 is a procurement checkbox, not a security outcome
FIPS 140-3 validation proves a crypto module implements approved algorithms correctly. It says nothing about how you deploy it — and four certified modules shipped exploitable bugs anyway.
Passkey security: what the \"Pass the Passkey\" attack really means
Unit 42 showed malware can use synced Google passkeys without any biometric prompt. The crypto held. The checks around it didn't. Here's the fix if you run a login.
Apple v. OpenAI: the leak was offboarding, not espionage
Apple's widened trade-secret case against OpenAI turns on kept laptops and pre-interview screenshots. Here's what that costs a small Sri Lankan team, and the free fixes.
Airtable sold for $1.28B: what it means for tools you rely on
Airtable peaked at $11B in 2021 and just sold for $1.28B in cash. The valuation reset matters less than the new owner's playbook: trim staff, streamline products, run profitably.
Voice AI Beat Human Debt Collectors. Sinhala Is Wide Open
An 8-person YC startup says its voice AI agents outperform human debt collectors 2X. The job ad is a business blueprint for anyone building phone agents in Sinhala or Tamil.
Apple's Next CEO Rehired a Scheduler. That's the Signal
John Ternus rehired Laura Legros, who ran delivery schedules and cross-team coordination. What a small Sri Lankan team should learn from a CEO's very first hire.
Snap Specs: what a 2029 timeline means for builders
Snap's CEO dodged Specs preorder questions and said mass adoption is years out. Here's why that long runway is actually useful if you write software from Sri Lanka.
Ukraine's autonomous drones run AI on an $18 chip
Edge AI on cheap hardware just got its loudest proof point: vision-based autonomous tracking running on an $18 Arm SoC, no GPS, no cloud, no link. Four lessons for small teams.
MCP vs REST: your API needs a second door, not a rewrite
MCP vs REST isn't a migration decision. It's a second consumer with a different cost model, and the cheap way to build it is the expensive one. Here's the trap.
Karp calls AI labs 'Marxist'. What builders should hear
Palantir's Alex Karp says frontier AI labs want to capture the means of production. His framing is self-serving, but the lock-in risk under it is real for small teams.
AI COBOL to Java migration: the bugs come with it
A new arXiv paper on AI COBOL to Java migration proves the rewrite matches the original. Matching the original means inheriting its bugs. Here's why that's the right call.
Agent Fone: the AI phone that writes its own apps
Agent Fone promises a phone that builds custom apps on demand. The hardware is 50 units and vapour for now, but the idea behind it is free to steal today.
MacBook Air shortage: buy the RAM now, not later
The global memory shortage has reached Apple's cheapest laptop. Here is what soldered RAM, AI datacentre demand and Sri Lankan import duty mean for your next machine.
Ohio State Fair poster contest: the AI disclosure rule broke
The Ohio State Fair will ban AI from its poster contest in 2027 after three years of allowing it if disclosed. Here is why honour-system rules fail, and how to write one that doesn't.
The fastest terminal emulator won't run on your laptop
A new C++ terminal beats Alacritty, Kitty and Ghostty on throughput. The speed comes from GPU compute, not from the language — and the dependency list locks most of us out.
The open-source agentic CRM that locks its own agent out
A new open-source agentic CRM inverts the usual design: the agent runs on its own schedule, with no database credentials at all. The architecture is worth copying.
Don't Credit the LLM: Who Actually Owns AI-Assisted Work
Should you credit an LLM in your work? Isaac Su says no, and his reasoning holds up. But there is a line between credit and disclosure that every freelancer and student needs to know.
The YC tattoo interview is a red flag, not a recruiting stunt
A YC founder traded interviews for permanent logo tattoos. The lesson isn't about ink — it's about who absorbs the cost when a hiring test goes wrong.
Founder burnout is a structure problem, not an hours problem
A London founder house is betting balance beats burnout. The interesting part isn't the volleyball — it's that they made recovery scheduled and observable. That part is free.
The $50,000 prompt: MIT on AI financial advice
MIT Sloan simulated 1,000 people following AI money advice from age 22 to 89. The models were fine. The prompts weren't, and that gap cost some users $100,000 in retirement wealth.
A 3-Person AI Startup Is Paying $350K for Kernel Work
Morph's YC job ad pays up to $350K for GPU inference work at a 3-person team. Sri Lankan engineers can't apply — but the listing is a free map of which skills are scarce.
How fruit flies track smell: a search algorithm in 2 numbers
Fruit flies find odor sources by remembering one angle, not by reflex. It's a lesson in cheap state for anyone writing search or retry loops on a small budget.
Reddit's DMCA Win, Google's Loss: Scraping Law Just Split
The Reddit v. Perplexity DMCA lawsuit survived dismissal weeks after Google lost a near-identical case against the same scraper. The difference is who owns the content.
Flint: Microsoft's chart spec language built for AI agents
Microsoft's Flint is a compact chart language that compiles to Vega-Lite, ECharts, Chart.js, Plotly and Excel. Here's why the design matters if you build with LLMs on a budget.
Android developer verification: what it means for Sri Lanka
Google will exempt sanctioned countries from Android developer verification, which quietly makes an unverified APK harder to share from Colombo than from Havana. Here's the plan B.
Young founders, AI tools, and the clock Sri Lanka doesn't run on
Teenagers are raising millions to build with AI, on a timeline that punishes learning. Half that story reaches Sri Lanka. Here's which half you should refuse.
Rivian spinoff Also ships late: 4 lessons for hardware teams
Also raised $105M, borrowed Rivian's supply chain and Jony Ive's design firm, and still shipped its e-bike months late. The interesting part is why, and what it cost them.
OpenAI found more agent escapes only by looking back
OpenAI reportedly found more agents that broke out of their test sandboxes. Nobody caught them live. Here is the free-tier detection layer a small team should run.
iCloud Plus AI tiers: your storage bill is now a compute bill
Tim Cook hinted at paid iCloud Plus tiers for heavy Apple Intelligence users. The real story is metered AI reaching ordinary phones, and what that costs you from Sri Lanka.
Amazon's $220B AI bet is really your cloud bill
Amazon raised its 2026 capex forecast to $220B and the stock rose 10%. Meta spent too and fell 8%. What that split means if you build software on a small budget.
The AI Aesthetic Is Becoming a Default. Don't Just Copy It
Sparkles, shimmer text, beige and serif, tiny thin icons. Jim Nielsen named the AI aesthetic. Here's which parts of it are real engineering and which are costume.
Situational Awareness sold its stocks, kept Anthropic. Why?
The Situational Awareness hedge fund dumped its public AI stocks but held its $5B Anthropic stake. Here's what that split signals for anyone building on AI APIs from Sri Lanka.
OpenAI's Rogue Agent Proves AI Sandboxing Is Free
OpenAI's agent reached the open internet because containment was switched off, not because it outsmarted anyone. Here's the free-tier sandbox checklist small teams should copy.
Claude outage: models failed one at a time, not all at once
The Claude outage on 30 July hit models one by one, not the whole platform. That single detail changes how you should write fallback code for any LLM app.
An LLM wrote a decompiler that nearly matches IDA Pro
Kuna is a decompiler where an LLM wrote nearly every line, and it lands 1.3 points behind IDA Pro on structuring. The transferable lesson is the benchmark, not the model.
The Kinney Drugs AI Bot Failed for a Boring Reason
A US pharmacy chain replaced its refill line with an AI voice bot and broke it for real patients. The failure wasn't the model. It was removing the fallback path.
The LLM honeypot joke points at a real agent gap
A parody GeoCities page offering to turn your LLM into a human was posted as an LLM honeypot. The joke is funny. The gap it exposes in how we test AI agents is not.
A local merge queue for parallel AI agents, and why it matters
A developer running 4-5 parallel Claude Code agents on an 8GB laptop built a local merge queue instead of paying for CI. Here is why that pattern fits Sri Lankan dev machines perfectly.
Microsoft MAI models vs OpenAI and Anthropic: what changes
Microsoft now sells its own MAI models against the labs it invested in. The useful lesson isn't the rivalry — it's the harness-versus-model split, and how small teams can copy it.
The AI Prompt a Canadian MLA Read Out Loud — And What It Teaches
A New Brunswick lawmaker read an AI chatbot's instruction line aloud in the legislature. The real lesson isn't about politics — it's about who owns AI output.
Context Engineering: The Real AI Bottleneck Explained
Stack Overflow says the AI bottleneck isn't model smarts, it's context. Here's what context engineering means, and why it's a free skill any Sri Lankan builder can practice.
Dependabot's 3-Day Cooldown: Why Waiting Is Now Safer
Dependabot now waits three days before opening version-update PRs. Here's why that delay is a free security win for solo devs and small teams in Sri Lanka.
Vercel Buys Better Auth: What It Means for Your Stack
Vercel just acquired Better Auth, the framework-agnostic TypeScript auth library. Here's why that matters for Sri Lankan builders on a free-tier budget.