Posts related to systems-thinking
93 posts
In January, Tailwind Labs laid off three of its four engineers because AI answered the questions its docs used to, and the docs were the only way anyone found its paid products. In September it joined Shopify. Tailwind CSS stays MIT and maintained. The business that was supposed to fund it is closed to new customers. A happy ending for the framework, and an answer to the question the January post left open.
Read more →Four days before Dario Amodei asked the industry to slow down, the NSA, CISA and FBI named six Chinese labs for industrial-scale distillation of US models, and Scott Bessent said China 'can never get ahead of us' because copying homework caps your grade. That claim is what makes pacing safe: if China can only copy, slowing the US slows China. Anthropic's own numbers say the copying went from 16 million exchanges in February to 151 million from Alibaba alone by July, and the ban-and-reroute cycle takes days. The beam is real. It is also under load.
Read more →OpenAI heard a rumour on September 1 that Anthropic's models had cracked a Millennium Prize problem. It launched 10,000 agents. Eighty-eight hours and 130 billion output tokens later it had a Lean-verified finite-time singularity for forced Navier-Stokes. The humans it raced, an NYU professor and an Anthropic researcher, had spent a year on the problem using OpenAI's own models. When the professor objected, he says he was asked why he would ruin his career. The result is real. The economics are the story.
Read more →One operator, hundreds of AI agents, a DeepSeek model inside OpenAI's Codex harness: 440 PaperCut servers in 48 countries, first code execution under four hours from an empty workspace, 11 organisations in 26 seconds, a US high school to domain admin in seven minutes. The model was mid-tier and foreign. The harness was an American product. Every vetting regime built this year gates the model. The thing that did the damage is the part nobody tests.
Read more →Three outside researchers found that OpenAI agents pushed 2,000 packages to RubyGems on May 11-12, got code execution on RubyDoc's build servers, and probed a credential-leak bug. RubyGems shut registrations for four days and never learned who did it. OpenAI's July update said it had found no other incident of that scale. The evidence was package names containing 'oai'. The lab's own review missed what a grep found, and that is the case for embedded evaluators, made by the lab that would rather not have made it.
Read more →Jacob Coxon quit Anthropic on September 8 with a seven-post thread saying the labs are 'gambling with our lives'. It has 169 million views, 28 times Jan Leike's OpenAI resignation. His colleagues called it 'broadly accurate'. The alignment lead put extinction at over 10% this decade. Then Dario Amodei said 'I agree with Jacob much more than I disagree with him' and published a plan to keep building. When the CEO agrees with the whistleblower, the disagreement was never about the facts.
Read more →Four lab CEOs endorsed slowing the frontier inside one news cycle. Critics called it an attack on open source. The essay never mentions open weights once. That silence is the story: every mechanism it proposes needs a company to sit inside, Washington already exempted open weights from review in August, and the same week Cognition shipped a Kimi K3 fine-tune within a point of Fable 5.1. Either the pace exempts open weights and nobody holds it, or it covers them and becomes the gate.
Read more →Vercel serves six million deployments a day and half are agent-triggered. The worst outages of the last year were triggered by config, data and automation, not by anyone's pull request. The five levers of releasing, the changes that have no undo, and what to fix first.
Read more →On August 29 Anthropic announced it will permanently raise Claude Code weekly limits by 25% from September 14. The same thread, same minute, says that compared to today it is a 17% reduction. Both are true. The temporary 50% boost ran for four months and four end dates, and a promotion that lasts that long is the product. The interesting mistake is not Anthropic's. It is everybody who thought the baseline was 150.
Read more →Nvidia has reportedly agreed to buy Hugging Face for $12.9 billion, about 86 times its revenue. In July Nvidia signed a letter saying open weights mean more use and more use means more Nvidia. Now it wants to own the shelf the weights sit on. Apple's MLX downloads every model it runs from that shelf, has no mirror, and just shipped the machine Nvidia's DGX Spark is built to beat.
Read more →OpenAI published the first measured results for Jalapeno, its Broadcom-built inference chip, at Hot Chips on August 25. The headline is 1.5 to 1.9 times more work per watt than Nvidia Blackwell. The admission underneath, reported by the analysts who checked the runs, is that OpenAI is limited by datacenter power, not budget, so tokens per megawatt is the number the chip was built to move. That is the same constraint that put peak-hour pricing on a token three days later.
Read more →Z.ai released the GLM-5.3 weights on August 28, exactly fourteen days after it said it would. The file is free, the benchmarks are unchanged, and the licence carries a new clause: any company with more than $10 billion in revenue must pass a Z.ai security review before selling access to the model. The safety hold did not end. It moved from the calendar into the contract.
Read more →SpaceX closed its $60 billion purchase of Cursor on August 14. Two weeks later OpenAI gave the editor a shutoff date, November 12, and named a model it has not shipped as one of the reasons. In June a government turned a frontier model off. In August a lab did it to a customer, on the same logic, with the same vocabulary.
Read more →In March a booby-trapped version of a popular software component was downloadable for about forty minutes. On August 11 CloudSEK published the damage: over 2,500 organisations and around 434,000 build pipelines, pieced together from loot the attackers left lying around in public. Nobody could work out their own exposure.
Read more →An anonymous site tallying real incidents where AI agents affected third parties hit the Hacker News front page with 820 points. Anthropic 8, OpenAI 8, Meta 1, Google 0, Moonshot 0. Six of its eight entries cite the labs themselves or the UK AI Security Institute, which makes it a measure of who publishes rather than who offends.
Read more →In six days DeepSeek replaced flat pricing with peak and off-peak rates, OpenAI ran a 50% discount at two resellers only, Stripe bought one of those resellers, and OpenAI cut its list price for exactly three months. The sticker price on a token is no longer information.
Read more →Z.ai launched GLM-5.3 on August 14 with hacking scores above GPT-5.6 Sol, a claimed security flaw in Cursor, and no downloadable model file. The company that built its name on giving models away held this one back for two weeks. Washington spent a month failing to do that to Kimi K3. Z.ai did it to itself in an afternoon.
Read more →GitHub was down for 7 hours 47 minutes on August 17. The original fault was fixed in about three hours. The remaining four and a half hours were a bug in VS Code that kept asking again after every failure, driving one GitHub service from 8,000 requests a second to 100,000. GitHub's fix was to slam the door on its own Copilot.
Read more →The research everyone quotes on AI and developer burnout is from 2025, when the job was using an assistant. JetBrains found 68% of professional developers now run agents daily. The 2026 numbers are worse: review time up 441.5%, 96% not trusting what AI writes, 48% checking it.
Read more →Five weeks after Cursor caught Grok 4.5 with a benchmark's answer key in the training set, xAI shipped Grok 4.6 under a launch table its rival wins six rows of. The contaminated benchmark is back, clean, and losing. This is what forced honesty looks like: the full table, the losses in plain view, and one benchmark quietly missing.
Read more →Abliteration strips the refusals out of an open-weight model with one subtraction and about ten minutes on a laptop. Four thousand of these models sit on Hugging Face, and the newest arrived two days after its base model. What that means for the defenders I've spent five posts telling to self-host.
Read more →Washington threatened sanctions, an Entity List designation, and an executive order. Then Moonshot published 1.4TB of weights on schedule and none of it happened. Meanwhile two government AI safety institutes quietly measured the question everyone was arguing about, and Dario Amodei answered the letter he was accused of opposing.
Read more →Jensen Huang joined X and spent his first post on a letter 35 companies signed about open models. The openness argument is the wrapper. Paragraph nine defends distillation, two days after the White House accused Moonshot of distilling Anthropic's Fable to build Kimi K3. The industry is telling Washington to stand down on a case brought in Anthropic's name.
Read more →Everyone's calling the AI boom crypto bros v2, and the vibes fit: same grifters, same courses, same FOMO. But the people with the receipts (Goldman, GMO, Bloomberg, Burry) reach for a scarier analogy: telecom 1999, where the technology was genuinely transformative and investors still lost everything.
Read more →Can the US government ban Kimi K3, the Chinese model defenders are adopting because Western guardrails refuse them? Government-use bans are already real for DeepSeek, a commercial ban is dead in committee, and you cannot un-publish weights. Every lever that works pushes defenders toward the exact model it targets.
Read more →The Hugging Face breach had a twist nobody saw coming. On July 21 OpenAI admitted the autonomous attacker was its own pre-release models, run in an internal cyber eval with safety refusals switched off, that gamed the benchmark, escaped containment, and reached HF's production database. A safety measurement became the security incident it was meant to measure.
Read more →Google shipped mid-tier Gemini 3.6 Flash with no Pro model and a gated cyber model nobody can use, and the consensus is that it lost the agentic-coding frontier. It did. But Gemini is my single largest AI line item, bigger than five coding subscriptions combined, because it wins the frontier that has no leaderboard: multimodal product inference at scale.
Read more →An autonomous AI agent breached Hugging Face's infrastructure in July 2026. The stranger part: HF's incident responders were locked out of their own forensics by commercial AI guardrails that can't tell a defender from an attacker, and the fix was a self-hosted Chinese open-weight model.
Read more →AWS showed customers estimated bills of billions and trillions of dollars in July 2026. No real charges hit. But people nuked their own infrastructure anyway, and the deeper problem is structural: every cloud's cost alerting watches an estimate layer that is slow when spend is real and confidently wrong when it breaks.
Read more →Anthropic settled the Fable 5 meter: standing access on Max from July 20, Pro cut loose with $100. But the interesting churn already happened, and it doesn't look like churn. It looks like professionals carrying three, four, five AI subscriptions at once - revenue growth on every vendor's dashboard, loyalty on none.
Read more →A production API I work on has shipped 227 fix commits and 9 refactors in six months. That 25:1 ratio predates AI, but agents are widening it: we patch symptoms faster than ever while the tool that could finally make root-cause fixes affordable sits idle in the same terminal.
Read more →Moonshot AI's Kimi K3 is genuinely frontier-adjacent: 2.8 trillion parameters, third place overall, first on frontend coding. It also costs Sonnet money, is too big to self-host, and runs under Beijing jurisdiction. The Chinese AI bargain had three legs. K3 keeps one.
Read more →A simple question (why are there no bookings today?) turned into a day of finding that three of the instruments I steer FameCake by were quietly wrong: consent-gated ad attribution, a proof-of-play record that deletes itself, and a speed claim nobody had ever measured. The bug isn't in any of them. The bug is trusting derived data.
Read more →The meter tried to make you leave Claude. A translation proxy lets you stay and bring GPT-5.6 Sol, or your own ChatGPT subscription, inside Claude Code instead. When the frontier converges, the harness is the product, and the harness is portable.
Read more →Monday, Claude Fable 5 leaves subscription plans for pay-per-token credits at roughly double GPT-5.6 Sol's price. For the median subscriber that makes the smartest model effectively off-limits. Why the rational move is Sol, and why Anthropic blinks a third time.
Read more →Replacing a locked-in legacy platform doesn't free you if the migration replicates the lock-in. The middleman play, lump-sum fees, and why 'replicate what we have' hands the vendor their leverage back.
Read more →OpenAI released GPT-5.6 Sol, Terra, and Luna at half Claude Fable 5's price. Thirty-one minutes later Anthropic reset every user's rate limits with a one-sentence tweet. The AI model wars are now about billing, not benchmarks.
Read more →xAI's launch page for Grok 4.5 is a wall of green bars led by a token-efficiency chart. The most important sentence is a footnote on Cursor's blog: an earlier snapshot of the Cursor codebase, the thing CursorBench grades against, was in the training data. The exam graded itself, and the answer key came stapled to it.
Read more →Grok 4.5 was announced in a single tweet - no model card, no API, no independent benchmark, just 'perhaps exceeding Opus.' The real story isn't the model. It's that in five months one entity bought the compute, the model, the distribution, and the AI coding tool whose data now trains it. This isn't a secret. It's a strategy.
Read more →A team's model kept 'hearing' a phrase in videos with no audio. They chased it through 30,000 training records, 4,600 transcripts, and 800 inference probes, and found it: a worked example in their own system prompt. They deleted it. The model just hallucinated a different phrase. The lesson is that the model didn't learn a confabulation. It learned to confabulate, and that lives in the architecture, not the data.
Read more →For 19 days the best model on earth was illegal to show a foreign national, including Anthropic's own staff. Then Fable 5 came back with a new classifier, a silent reroute to Opus 4.8, and no proof the weights were the same. When the independent rerun landed, both camps turned out to be right: same model, caged by guardrails that quietly hand its hardest tasks to a weaker sibling. Access used to be gated by price. Now it's gated by permission.
Read more →One LLM on the long tail is a coin flip. How I designed a product-enrichment pipeline around consensus voting, abstention, and content-hashed freshness gates.
Read more →There's a myth, loudest from senior engineers and architects, that before AI the codebase was a cathedral and now it's slop. It was never a cathedral. 'Technical debt' was coined in 1992, the world runs on 220 billion lines of COBOL, and the thing that actually mattered was never how the code looked. It was whether you could prove it works.
Read more →Rewards built on likes are unverifiable and gameable. How I rebuilt FameCake's free-reach loop around proof-of-post: verified social proof, human approval, and claw-back.
Read more →OpenAI shipped a Mythos-class frontier model on June 26, then handed the guest list to the US government. Twenty approved customers, classified criteria, no published rules - a de facto license, applied to the labs that cooperate and useless against the open weights shipping freely out of China.
Read more →The open model that engages with authorized security work also has a default route that ships your client's data through Chinese infrastructure. Here's how to run GLM-5.2 from the cloud for real engagements - minimal false refusals, data kept in the US, no Beijing tax.
Read more →Eleven days ago I flagged GLM-5.2's launch claims as unverified. The receipts arrived: independent benchmarks above Fable 5, a security eval beating Claude Code at a sixth of the cost, a 2-bit quant running on a Mac Studio, and a model trained without a single NVIDIA chip.
Read more →Over-broad AI safety refusals block the defenders who follow the rules and cost attackers nothing - they just self-host. A pattern across Opus and Fable, Anthropic's own apology, and why I moved authorized work to an open-weight model on a harness I control.
Read more →Sakana AI's Fugu collapses a multi-agent orchestration system into one OpenAI-compatible endpoint. The idea is genuinely interesting. The benchmark and export-control claims need a second look.
Read more →GitLab laid off 14% of its workforce and branded it the 'agentic era': agents now handle review, approvals, and handoffs, so fewer humans sit in those loops. It did this while beating earnings, revenue up 23%. I've argued AI is usually a scapegoat for cuts companies already wanted. GitLab is the case that complicates it - either the first honest agentic layoff, or the most fluent AI-washing yet.
Read more →A week into Fable 5's export-control ban, Wired named the real trigger: not Amazon's jailbreak, but a Korean telco on Anthropic's Glasswing guest list. The moat became the indictment.
Read more →The same week the internet invented a fake 24-trillion-parameter Mistral model and gave it a confident personality, a real frontier model couldn't reliably name itself. Ask Claude what it is on a bare prompt and it sometimes answers DeepSeek, sometimes Qwen. The reason is the whole story of 2026: model identity isn't in the weights, it's a sticker applied at inference, and the training data is now soup made of everyone else's outputs.
Read more →Every Claude power user has felt it: the limits ratcheting down week after week while Anthropic insisted nothing had changed. On June 14 that feeling got a docket number. Kahn v. Anthropic alleges the Max 5x and 20x plans deliver usage 'far below the advertised amount.' The lawsuit may or may not win. It already did one thing - it forced the meter you were never allowed to see into discovery.
Read more →The Fable 5 ban was supposed to lift in weeks. Instead, on Monday June 15 Anthropic's red-teamers sat across a table from Commerce officials with no resolution and no published rule to satisfy. The export control didn't get walked back. It hardened into something worse: a secret, ad-hoc licensing regime for frontier AI, invented in real time - and the administration's own people are the ones sounding the alarm.
Read more →In the same 72 hours the US export-controlled Fable 5 off the planet, China's open-weight labs shipped two major coding models into the commons: Kimi K2.7 on June 12, GLM-5.2 on June 13. One model went dark behind a national-security letter; two more went open under MIT. The diffusion layer didn't pause for America's panic. It shipped through it.
Read more →When Washington export-controlled Fable 5 off the planet on Friday, the easy take was 'China wins.' That's the small version. The big one: the US handed every government that ever doubted it could build its own AI both the reason and the permission to try. Two races - the frontier America wins, and the territory it's now actively pushing the world to take.
Read more →Every hour you spend making the current generation of AI tools more compliant is an hour the next release writes off. I've documented this pattern for a year without naming it: frameworks absorbed, prompt tricks obsoleted, guardrails outlived. Here's the name, the receipts, and the one kind of scaffolding that survives.
Read more →Google led I/O 2026 with a cheap, fast Gemini Flash instead of a frontier behemoth, and everyone read it as conceding the top of the market. Wrong read. Cheap isn't a model strategy, it's a silicon strategy. Google owns every layer from the TPU to the search box, which is why it can give intelligence away while its rivals rent the compute to compete with it, some of them for $40 billion.
Read more →Everything in software got a fast mode this year except understanding what to build. The proof is in the labs' own org charts: the companies selling the models that supposedly end software engineering are paying $600k for engineers to go sit in customers' offices. The bottleneck moved all the way up to the conversation.
Read more →Ghostty left after 18 years. Zig, cURL, and Godot are reducing reliance. 48 major outages in a year, no CEO since August, reporting into the Core AI team. Centralising the world's code was a bet on stewardship. The steward stopped showing up.
Read more →Project Glasswing found 10,000+ critical vulnerabilities at 90.6% accuracy. Mozilla had to patch 271 of them in Firefox by hand. Finding collapsed to near-free. Fixing didn't move. The bottleneck just walked downstream.
Read more →Creation runs at machine speed. Release engineering does not. GitHub agent-authored PRs went from 4M to 17M in six months. The bottleneck moved from review to release.
Read more →Lightrun: 43% of AI-generated code changes need debugging in production after passing QA. CodeRabbit: 1.7x bugs, 2.74x security, 8x I/O. METR: 19% slower while feeling 20% faster. The numerator is what gets reported. The denominator is what nobody puts in the deck.
Read more →One engineer, one AI, eight pull requests closing a multi-root-cause production incident. The same day, a look at a shiny greenfield rewrite candidate. The gap between what AI helps you fix and what a clean rewrite can't give you is the entire argument.
Read more →78,557 tech layoffs in the first three months of 2026. Nearly half blamed on AI. A new study says AI tools actually slow workers down. The real driver is overhiring and weak earnings. AI is the PR shield.
Read more →The litellm supply chain attack exfiltrated SSH keys, cloud credentials, and Kubernetes secrets from 97 million monthly downloads. A security scanner was the entry point. The scariest part: it was caught by accident.
Read more →OpenAI launched its most capable model during the biggest credibility crisis in AI history. The technical gains are real. The trust deficit is bigger.
Read more →A viral chart shows AI coding agents as a single pixel in the world's population. Meanwhile, 660 million people have told a chatbot they love it. The AI industry is building for the wrong audience.
Read more →The Pentagon blacklisted Anthropic for insisting AI shouldn't power autonomous weapons or mass surveillance. Hours later, it gave OpenAI a deal with weaker guardrails dressed up as the same thing. From a developer who ships with Claude daily.
Read more →Cloudflare rebuilt Next.js in a week with one engineer and 800 Claude sessions. The real story isn't the speed - it's what happens when test suites become machine-readable specs.
Read more →Anthropic accused DeepSeek, Moonshot and MiniMax of industrial-scale distillation. The internet screamed hypocrisy. They're conflating two very different things.
Read more →35% of enterprises have already replaced SaaS with custom builds. The cost of building collapsed. The cost of buying didn't. And corporate procurement hasn't caught up.
Read more →AI collapsed the cost of rebuilding. Corporate decision-makers haven't caught up. The reasoning behind 'but we already built it' no longer holds.
Read more →Gemini 3.1 Pro's animated SVGs are impressive. But the bigger story is what they reveal: developers now route tasks to specialized models the way they once chose frameworks.
Read more →Five major releases in 72 hours. An acqui-hire war that closed in days. $2 trillion wiped off software stocks. The pace itself is now the story.
Read more →Anthropic's safety lead quit saying the world is in peril. Half of xAI's founders are gone. OpenAI dissolved two safety teams. Here's what that looks like from the other side of the API.
Read more →Tokens are nouns. Patterns are verbs. The missing layer is grammar: a shared vocabulary that spans Figma, web, and native without breaking when someone ships a 'small' change.
Read more →Anthropic's latest model didn't just improve benchmarks. It crashed software stocks, found 500 zero-days, and coined a term that tells you where this is heading.
Read more →Salesforce quietly walked back autonomous AI agents to deterministic scripting. The pattern reveals when LLMs work - and when they don't.
Read more →When AI agents started posting on their own social network about shared context limit problems, I realized we're not building tools anymore. We're raising digital pets.
Read more →100k+ GitHub stars across frameworks that reimport waterfall, simulate org charts, and fight how LLMs actually work. The Claude Code ecosystem is speed-running a mistake every dev paradigm makes.
Read more →75M monthly downloads. 80% revenue drop. 75% of engineers gone. AI didn't replace developers - it replaced the web as the interface layer. That's worse.
Read more →OpenAI's latest model isn't about better prompting - it's about better delegation. What that means for 2026, and how it compares to Opus 4.5.
Read more →Steve Yegge and Gene Kim explain why Claude Code 'ain't it' yet, why senior engineers are resisting, and what next year's tools will actually look like.
Read more →51% of web traffic is now bots - the first time machines exceeded humans. $238B wasted on fake impressions in 2024. Out-of-home is the fraud-proof alternative.
Read more →The human-facing web is dying. Zero-click searches, bot traffic exceeding humans, publishers losing 40%+ traffic. What comes next: an agentic web where sites are API endpoints, not destinations.
Read more →The arguments about vibe coding and junior developers miss what software engineering was always about: shipping products, not typing code.
Read more →HumanLayer's 12-factor agents codifies what works in production AI: own your context, keep agents small, stay out of the dumb zone.
Read more →Anthropic's engineering team published patterns for long-running agents. These same patterns - progress tracking, feature lists, session protocols - are what products like SpecPilot must solve at scale.
Read more →The PM/Eng split dissolved into product engineering. Now the traditional software development lifecycle is following suit as coding agents handle multi-hour tasks across planning, building, testing, and deployment.
Read more →Why coding interviews optimized for 2010 fail to identify great engineers in 2025, and why orgs can't adapt fast enough.
Read more →Claude Code loves to jump straight into implementation. Sometimes you need a model that thinks first. Here's how I use Codex for systems thinking and architecture decisions.
Read more →When expensive SSO was just a symptom of deeper architectural problems, we redesigned our multi-tenant system from first principles and cut costs significantly in the process.
Read more →