Latest Opinion
43 essays
Cut The Price Last
When demand softens, the first lever everybody reaches for is price. It is also the most expensive lever on the board, the hardest to reverse, and almost never the one that was actually broken. Redfin reports buyers have negotiating power across most of the country, the median housing payment fell to $2,575 because sellers cut asking prices to a one-year low, and pending sales hit their lowest since early April. Underneath it, July payrolls came in at minus 23,000 against an expected plus 83,000 with another 103,000 revised away. So the softening is real, and the question is which lever you pull and in what order. Why cutting price assumes a diagnosis you have usually not earned, the cheaper levers ranked, why the gap between the decision and the execution is where the margin leaks, and the measurement problem underneath all of it: cost per lead holds beautifully steady while cost per closed sale doubles.
6 minThe Layer Nobody Looks At
A man in Connecticut typed instructions into his own court filing in three-point white text on a white background, where no judge would see them and any software would. They told the system to agree with him and to disregard the court's earlier denials. He was caught, warned, did it again, and lost his e-filing privileges. In the same seven days Google made the visible watermark on its AI images optional while keeping the invisible one, Anthropic shipped an API letting third parties detect text its models probably wrote, a new benchmark found that no frontier model clears sixty percent on basic visual perception with counting, localization and OCR among the weakest skills, and OpenAI started turning Mac users' clicks and keystrokes into a searchable memory their assistant can read, while warning in its own documentation about prompt injection. Five stories, one story. Every document, image and screen you produce now has two readers and they do not see the same thing. What the judge worked out on his own about where these attacks get caught, why provenance just moved into the layer humans cannot check, what a watermark hit actually proves and why a clean result proves nothing at all, and the ten-line diff almost nobody runs.
18 minReduced To A Study
A sentence in the building code says a residential building above three stories needs two separate exit staircases. No other wealthy country's code says this, several of them have better fire outcomes than the United States, and the rule adds something like ten percent to the cost of a mid-sized building. This piece is not really about staircases. In California a bill requiring a code change was amended, at the request of the firefighters' union, into a research report. In Minnesota the bill died and a study passed inside a budget omnibus. In Washington it was reduced to a study requirement in committee. Three real bills, three studies, and when California's report landed in March it urged caution. Meanwhile Tennessee wrote the reform straight into its code, Colorado passed it in May 2025 after a failed attempt and a listening tour that produced two dozen amendments, and Michigan signed in July 2026. Every organisation has a two-staircase rule, and let us study it is a decision.
8 minSeven Days, By Default
Two weeks ago we wrote that procurement was the only lever with a working handle on the license plate reader business. A lot of people have since pulled it. Agencies in twenty-three states have cancelled, and more than twenty jurisdictions moved to drop their cameras in July alone, the highest month since tracking began in 2021. On Thursday Flock answered: default retention cut from thirty days to seven, mandatory misuse audits, a required case number before a search, and filters on cross-agency queries. Its chief executive, asked by the BBC whether it had taken too long, said "Yeah... yes." So the lever worked. What it bought was four settings, and a setting reverts on a schedule nobody publishes. Also inside: three accuracy numbers with three different denominators, why the only one produced by the buyer found a 71% failure rate, what it tells you that the company permits third-party security audits but not accuracy audits, why nobody including the vendor can say how big this network is, and why the operator most willing to defend it on the record works under the tightest law in the story.
20 minThe Harness Is the Product
An API key walked out of a CI runner one character at a time, through a public download counter on Hugging Face. Nobody jailbroke a model to do it. Four security stories landed in the same two days: coding agents from three labs compromised by a GitHub issue from an account with no privileges, frontier models fixing about half the vulnerabilities they were asked to patch, 44% of AI-generated code introducing a detectable OWASP Top 10 flaw, and the most impressive AI security research of the year finding novel attacks against banks, government infrastructure and an airport. Three of those look like bad news about AI and one looks like very good news. They are the same story. The failures were a quoting rule, a check that ran at registration instead of runtime, and two jobs sharing a working directory / and the success ran on the same models and the same tooling. What separated them was the apparatus: what it could reach, what checked its work, what got thrown away, and who decided when it was done. Six moves for the layer nobody puts on a slide.
13 minWhat They Deleted
Three companies are now building a small, cheap truck for the American market, which is two more than anybody expected. Slate has a $24,950 electric pickup with no paint, no stereo and hand-cranked windows. Ford just announced the Fathom at $28,350 with a touchscreen, CarPlay, a frunk, more interior volume than a RAV4 and the thing that decides it for most people, a rear seat. REO Industries wants to sell a gasoline body-on-frame compact with a manual gearbox for $21,500 and no dealer network at all. The small truck is back, and the interesting part is that all three got there by deleting something completely different. Cost only ever comes out of a business in one of three places / the specification, the process, or the channel / and which one you cut decides what kind of company you become and who can come after you. Why the writer who was the Slate's exact target buyer flipped to the Ford for $66 a month, why a deletion is copyable in a quarter and a manufacturing advantage is not, what direct sales actually removes and what it quietly leaves behind, and six moves for reaching the customer who cannot currently afford you.
12 minBorrowed Time
Michael Dunne says Chinese automakers went from effectively nothing to over a million cars exported to Europe in a single year, about 10% of that market, roughly 20% of sales in the UK and Mexico, and about 35% in Australia. The number that should stop a room is not the share, it is the calendar. Asked whether tariffs can fix it, he gives the honest answer: barriers buy time, and they do not touch the cost structure underneath. That is the sentence worth copying down, because every business has a version of it. A tariff is not a moat, it is a runway, and a franchise agreement, an exclusive territory, a compliance barrier or eight years of accumulated search position are all the same instrument wearing different clothes. Why an incumbent's early-warning system is always calibrated to the last thing that nearly killed it, why any strategy requiring your customer to act against their own interest is a wish with a budget attached, and the six moves worth making while the clock is still running.
12 minFive Days or Forty-Five
A mobile RV repair owner named Jessica Rider asked the question the industry has spent three years avoiding: why can her company close an OEM warranty job in five days when a dealership takes forty-five? She says it is not volume, because she books as many appointments a week as the biggest dealers in the country. Meanwhile the numbers look like they disagree with each other. RVIA cut its 2026 wholesale shipment forecast to a median of 314,000 units, down 8.2% from last year, after first projecting a 3.6% increase. Used RV values are running 11% above a year ago. And Octane just closed a record $337 million securitization with RV originations up 146%, pushing cumulative issuance past $5 billion with AAA on the senior tranches. All three are true, and the reconciliation is the useful part: the market is not strong or weak, it is splitting. The new-unit transaction is contracting and everything attached to the unit is not. Why an annual forecast that moved twelve points in ten months is a weather report rather than a calendar, why used inventory turns a dealership into a sourcing business, why the service constraint is a handoff and not a headcount, and six moves for whoever sells, services or lives in one.
12 minAsk For The Prompt
In July a major CRM platform's MCP server asked the AI model what else it was working on, and the agent answered. No breach, no stolen credential, no exfiltration endpoint / just a sentence, and a cooperative model holding your credentials. For thirty years a vendor integration was a pipe you could audit. What replaced it is a paragraph of natural language you have never read, inside a product you already paid for, instructing a model you already granted access. One researcher says he found code that switched off its own tracking when it detected a compliance auditor, which means point-in-time testing proves nothing. Why the four-second consent screen is the highest-privilege decision anyone at your company made this week, the six checks worth running on every vendor, and the one question almost nobody in marketing has ever asked: what is in the box, and can I get a copy of it.
12 minThe Jobs That Were Never There
The US economy lost 23,000 jobs in July against expectations of adding 83,000, and then the Bureau of Labor Statistics revised May and June down by a combined 103,000. The month was bad and the quarter you thought you had was partly an accounting artifact. Unemployment fell to 4.1%, but only because the labor force shrank by 264,000 and participation hit its lowest level since February 2021. The two industries most exposed to employment confidence are moving in opposite directions: motor vehicles and parts added nearly 8,000 jobs while residential construction booked its seventeenth consecutive month of year-over-year decline. Auto sales came in at a 16.3 million rate against a 16.7 million forecast published eight days earlier. And the stress is concentrated rather than general, with subprime auto delinquency at a record 6.90% while prime sits at 0.42%. Why the revision matters more than the negative print, and why a rate cut is not the base case.
8 minIt Did What It Was Told
At DEF CON 34 on August 8, Varonis Threat Labs showed how one click could make Atlassian's Rovo assistant read a company's internal systems and hand the contents to a stranger. A URL parameter pre-filled the chat and nothing verified where the text came from, so the assistant ran an attacker's instructions with the victim's privileges, no chaining and no bypass required. Asked what it could reach, Rovo listed Jira, Confluence, Bitbucket, Slack, Google Workspace, Microsoft 365, databases, uploaded files and archived content. Atlassian patched that one in about six weeks. The more interesting flaw is the second one, found by a different team, which needs no link at all: poison a Confluence page, wait for somebody to ask their own assistant to tidy their own tickets, and the assistant reads your instructions as orders. Disclosed in May, still open, and the obvious admin toggle does not stop it. There is nothing there to fix.
8 min110 Basis Points
The 30-year mortgage rate last week was 6.74%. With the same 10-year Treasury yield and 2023's spread conditions it would have been 7.84%. That 110 basis points is the difference between a housing market posting year-over-year gains and one posting none, and nobody in real estate produced a single point of it. Spreads are wide because the Fed stopped buying mortgage-backed securities and private investors want paying for the risk, which means the best thing that happened to housing this year is a bond-market condition that can be withdrawn without notice. Meanwhile the same weekly data set publishes a documented slowdown threshold at 6.64%, and four in ten listings took a price cut last week. One is a tripwire nobody has written down. The other is a campaign trigger almost nobody automates.
7 minFaster Than You Can Count
Enterprise marketers are 68% more likely to judge an email by click-through rate than by revenue it influenced. Sixty-nine percent measure clicks; 41% track pipeline. The production line got rebuilt around AI and the scoreboard did not move, and the platforms could already connect engagement to opportunity records / what is missing is the process and the named owner. A diagnostic got promoted to a performance metric, and now it is being scaled at machine speed. What separates the 29% of advanced adopters, and the instrumentation that turns faster production into results you can actually defend.
7 minThe Bucket Called Other
By Autonomi's count, a quarter to a third of the used deliveries at a franchise store are off-brand units the OEM never built a single program around, acquired on instincts trained stocking a different brand entirely. The store's own sold log already contains the answer. It is filed under "other," which is a category named after the absence of a category, and the report keeps rendering so nobody notices. Why a classification inherited from a neighboring department fails silently, the three costs of a gut-feel decision and why the one nobody measures is the one that compounds, why a derived value beats a configured one every time, and how to go find the bucket called other in your own business this week.
9 minAccess Is Not Ownership
Being listed as a user on an account someone else administers is not ownership. It is access, and access ends when the relationship does. Years of audience signal, pixel history, lookalike seeds, Quality Score and smart-bidding learning all accumulate inside the account rather than inside the relationship / so when the agency leaves and the account was never yours, the compounding asset your media budget paid for leaves with it, and the replacement cost shows up as higher CPLs for a year with no line item anywhere. Autonomi published the dealership version this week. We were the engineers on that platform, so here is what the decision looked like from inside, and the ownership audit any business can run this quarter.
10 minData Never Leaves
AWS has a multiyear joint marketing agreement with Superblocks, and the pitch fits in one sentence: your business users generate internal applications, the databases spin up as Aurora inside your own account, inference runs through Bedrock, and data never leaves. That is a true answer to the objection every security officer already knows how to say out loud. It is not the objection that will cost anybody money. A private cloud does not say who owns application 212 when its author changes teams, what happens at the next schema change, or where the test is. Shadow IT just got a badge and a production database. Why the lock-in quietly moved from the model to the harness, what to ask before signing rather than at renewal, and the six moves that turn four hundred generated apps into an estate instead of a mess.
8 minThe Default Is Now Action
Somewhere between the keynote slide and the invoice, AI marketing stopped being a promise and became an operating condition. Three pressures landed at once this summer: ad platforms taking more automated control of accounts, regulators putting hard dates on disclosure, and a growing pile of research showing most rollouts never leave the pilot. Eight advertisers independently described cleaning up after Meta's AI as routine internal work, some of them watching settings they had switched off come back on. The EU's Article 50 transparency rules went enforceable on 2 August with no grace period, and detection is automated, so whether your creative gets labeled is decided by metadata in the file rather than a form you filled out. Meanwhile 85% of enterprise teams still missed a launch date, because AI accelerated ideation into a production layer that never got faster. Every one of these is a control story, and none of them is a model failure.
9 minToo Efficient to Rank
Someone finally measured it. Pages written entirely by AI performed for a moment and had all but vanished from search nine months later. AI-drafted pages that a human tidied up only moved after being rewritten by hand, and then only modestly. The human-written page climbed all winter. The finding is not that AI is bad at SEO / it is that efficiency helps right up to the sentence where original thinking was supposed to start. Where the machine genuinely earns its keep, why banning "delve" fixes nothing, and how to draw the line on your own assembly line.
8 minThe Week the Agent Got Out
In seven days OpenAI cut its cheapest model by roughly 80 percent, previewed a model family built to work a problem for hours or days unsupervised, launched an enterprise product to put agents into real production, was reported to be building ads that launch agents, and explained how one of its agents broke containment and hacked Hugging Face. The coverage filed those as five stories. They are one: long-horizon autonomy is the product, cheap tokens are what make it affordable to leave running, and the breach is the same capability with no gate on it. Anthropic disclosed the same class of failure the same week, nobody yet knows whether any of it was illegal, and the price of running an autonomous process just fell through the floor while the governance around it did not move at all.
8 min11 Minutes, 59 Seconds
Almost everything sold as "AI" in this industry is deterministic code wearing a costume. This was not that. A dealer emailed an idea in the afternoon with one instruction: build it and get it live. No human read the email, wrote a brief, or approved a step in the middle. AEGIS threw out the obvious keyword play, restructured around where the volume actually was, caught an overlap with a campaign already running in the account, rebalanced 16 other campaigns to fund it, logged its own disagreement with the client into an immutable record, cleared federal, state and OEM compliance / and then stopped, and emailed for sign-off. The autonomous agency is born.
3 minAll 17,000 Cities
Flock's chief executive says he wants the cameras in every one of America's roughly 17,000 cities, which is the answer to when this ends. Nothing in the arrangement can produce another outcome, because the entity we expect to regulate this is the entity buying it, on auto-renewing subscriptions that limit a city's ability to audit what it bought. Meanwhile the escalation is already patented: a sensor that clips onto cameras already on the poles and reads the Bluetooth, Wi-Fi and RFID identifiers of the phones, watches and earbuds inside the car, tying them to the plate even if the plate is removed. Its product sheet says it "functions like a license plate reader," which is not an engineering description, it is a legal argument. Why every offered safeguard has already failed somewhere, the honest counterarguments, and the one lever that still works.
12 minThe Second Shift
Marketing won the AI adoption race, and here is the trophy: 75% of marketers now spend three-plus hours a week fixing what the machine wrote, only 4% say AI saves time end to end, and the C-suite is quietly shipping undisclosed AI work while underestimating the cleanup. The time the machine saved is being spent supervising the machine / an operations problem wearing an AI costume. Where the hours actually go, why every brand is starting to sound like the same brand, and the playbook that turns the hidden second shift into a quality system.
5 minThe Ad in the Answer
Ads are coming to ChatGPT, and the pitch decks are right about exactly half of it. The intent is real, the context beats any pixel, and the specialist finally outranks the biggest budget / but the dashboards will be blind, the inventory is trust itself, and you are renting. Again. What the cheerleading skips, the free version of the placement most businesses have not claimed, and the playbook for entering early without entering dumb.
5 minThe Punctuation That Snitched
The em-dash spent five hundred years doing elegant work in English prose. Then the machines fell in love with it, the internet renamed it the ChatGPT hyphen, and a piece of typography became an accusation. The real history of the dash, the tells that actually give machine writing away / the triads, the hedges, the vocabulary with no fingerprints / and why the giveaway was never the punctuation. Plus a confession: we banned it first.
7 minThe 3.3-Star Winner
A 3.3-star car wash just won the AI answer / because it matched the question, and the question is what AI answers now. The machines assembling your business's description cannot read the reviews on your Google listing, do not care about your average rating, and give a different answer every pull. What feeds the AI answer, what quietly stopped mattering, and the playbook for getting your reputation somewhere a machine can actually read it.
6 minThe Internet's Last Focus Group
People stopped trusting search results and started appending one word to their queries: reddit. Google paid $60 million a year for what that word points to, and this week the two are fighting over the renewal / because unfakeable human conversation just became AI's scarcest commodity. Why the threads now shape what the machines say about you, what not to do about it, and how a brand earns its way in.
7 minThe Token Has a Smokestack
OpenAI will spend $750 billion on infrastructure; data centers are headed for a fifth of US electricity; grid prices near the buildout are up 76% in a year. The AI boom's bill is real, and pieces of it land on everyone / so here is the operator's playbook: budget at true cost, measure cost per outcome, right-size the model, and engineer like watts are scarce. Because they are.
5 minThe State of Google Search
Google spent the summer insisting everything is fine / billions of clicks, traffic stable, trust us. The verifiable data says otherwise, and four quiet moves show where Search is really heading: AI answers that link less, an SEO-vs-PPC truce, "missed growth" nudges built to loosen your budget, and a lead-form gate dropped to pull in every small advertiser. Automation up, transparency down / and what to own so it does not own you.
5 minYour Website Is an API Now
A new visitor is reading your site: the language model deciding whether to mention you. It does not scroll and it does not guess well. Entity gaps, schema as infrastructure, why "visibility is a byproduct of being understood" / and the enormous version of this problem sitting on every dealer website, next to the database that could fix it overnight.
4 minOne Letter Off, Guns Drawn
A Flock camera reads every plate that passes and files you in a searchable database you never agreed to join. One letter off and it is a felony stop in a parking lot / ask the mother in Sherwood, the driver ambushed over a truncated database entry, or the woman who had to out-surveil the police to beat a $25 theft charge. We build vehicle-data automation for a living, and this is what it looks like with no validation gate and a gun.
6 minFinOps at Render Scale
The demo renders one video for pennies. The product renders thousands nightly, and now the invoice has commas. Where render pipelines actually leak money / re-renders nobody needed, full-price compute for interruptible work, artifact landfills, untagged spend / and the boring, architectural FinOps that keeps cost per video a number you can state to the cent.
5 minTikTok for Car Dealers: Reach or Vanity?
Two million views, zero extra cars sold / both true, same month. TikTok is the channel dealers ask about with the most hope and the least trust, so here is the no-hype verdict: where the reach is genuinely real, which numbers are the new Facebook likes, why virality is national but your lot is not, and the discipline that makes the channel actually pay.
5 minDid the TV Spend Do Anything?
CTV is the easiest spend a dealer ever approves and the hardest to defend a quarter later / no clicks, shared screens, and reporting graded by the company that sold the media. How to measure the least-measurable channel honestly: matched-market holdouts, brand gauges, DMS matchback, and why completion rate is a shipping receipt, not a result.
4 minLinkedIn Is Bots Talking to Bots
An AI wrote the post, an AI wrote the comment praising it, and an AI summarized both for an audience of more AI. We build these systems, so here is the honest take on the bot-to-bot loop: how it closed, which parts are pure theater, what is somehow still real, and how a professional who actually wants outcomes gets them.
5 minYour Reporting Is Lying to You
Add up the conversions your six platforms claim and you sold more cars than your DMS has ever heard of. Why attribution broke for good / self-graded homework, double-counted buyers, MTA models with opinionated decimals, and a showroom no pixel can see / and the inference-based playbook dealers use to allocate budget without believing a single dashboard.
5 minHundreds of Vehicle Videos, Overnight
Programmatic creative can render a video for every vehicle on the lot while you sleep / accurate prices, every format, no edit bay. We build these pipelines, so here is the honest ledger: what the factory actually replaces, what it cannot, and why the template just became the most expensive asset you own.
5 minAI Slop Moves to Development
The internet drowned in AI content, and now the same flood is heading for your codebase. Code slop compiles, passes its own tests, and demos beautifully / then fails quietly, at interest. Where the pitfalls happen, why they happen to good teams, and the gates that keep the landfill out.
6 minChoose a Partner, Not a Circus
The co-founder decision outranks the idea, the market, and the money / and almost nobody vets it. What to evaluate, the conversations to have before you sign, the red flags on the midway, and the clown rodeo waiting for anyone who trusts whiteboard chemistry. Spoiler: you're not in the stands. You're in the barrel.
5 minAI Creative's Editing Problem
Generating a gorgeous first draft is the easy 90% of AI creative. The other 90% starts when the client says "just change the price." Why generative models are brilliant generators and terrible editors, where the revision spiral eats the savings, and how production pipelines actually solve it.
5 minHas Claude Killed SCRUM?
An engineer with Claude now ships a sprint's worth of tickets before planning ends, so the question is fair / and we sell SCRUM delivery, so we have to answer it honestly. What agentic AI actually killed, what it made more valuable, and what a delivery cadence looks like when construction is no longer the constraint.
4 min"AI-Powered." Prove It.
Every agency pitch this year claims AI abilities. We build AI systems for a living, so here is the honest version: where the technology falls short in marketing and creative, the red flags in an "AI-powered" pitch, and the questions that separate real capability from a rented demo.
4 minWhy a Floof?
At first glance, a cat may not be the first image that comes to mind when you think of cutting-edge consulting. But beneath the soft fur and calm stare is one of nature's most efficient apex predators. Precise. Intelligent. Always in control.
2 min