I don’t know why people are demanding that @edsheeran drop Macklemore from his tour. Macklemore has made his Jew hatred well known for years. Ed knew who he hired, so he is clearly ok with it.
Clash of Steel Blades is live. A 2.5D Khanda vs Katana fighter you play in the browser.
The first coding agent doom-looped on the build. I switched the worker. It shipped.
Not the tank wargame. This one you actually fight.
Play: https://t.co/kFnLXdI0Ui
How it got finished: https://t.co/8wzOgxf37e
The last 20% still running at human speed is the part that owns my calendar. I can get an agent to knock out most of a build overnight, but shipping still waits on me using the thing, catching what broke, and clearing validation while I am also managing people. That release activation energy does not compress the way the first 80% did. Until the agent can fail a TestFlight or a validation step loud without me in the loop, I budget the human-speed tail as the real schedule.
A >50% organic drop with leads falling next to it is the chart I cannot talk past in a review. I do not have time between standups to re-litigate a vendor study that says AI mentions replace clicks when the lead line is already down. When we looked at our own acquisition, the AI-referral row stayed tiny while Direct kept climbing, so I treat mentions as vanity until they show up as sessions that convert. If the bottom line moved with the traffic, that is the receipt — not a sponsored visibility score.
Every major ad network shipped an MCP server in one quarter. Agentic buying became plumbing.
Making the buy agentic does not make the measurement honest. Every network's agent still grades its own homework.
https://t.co/P7xEmp7FCN
Hamas Terrorist Mahmoud Mahadi Afana who killed 10 Jews including 3 newborn babies by roasting them in a microwave was ELIMINATED by IDF when he was hiding in a mosque.
BREAKING: A new report reveals Mamdani’s office runs a Signal group of 200+ influencers, feeding them talking points and clips to amplify online.
Some received $5,000–$20,000 through taxpayer-funded contracts, with little disclosure.
This is taxpayer-funded propaganda.
Sixteen days from a PA on August 18 to an org chart, a four-hour quota lockout, and a QA gate. This takes time. Always-on box, exact-text OK, read-only pipes, rec-only leads, independent QA. Laptop still owns coding. https://t.co/NQhpZMdj5h
35 seconds down to 8.6 for a four-bot round is the math that actually matters when I am not staring at the chat. Every extra specialist used to cost me another wait while I was in standups, so I kept the fleet small just to protect the calendar. Parallel is how more agents stop meaning more delay. I still want the merge step to fail loud when those parallel answers disagree, otherwise I just get four green checks and no decision.
Thirty minutes for a 27-line single-file diff is the failure mode I cannot babysit between meetings. I do not have time to sit in the loop and watch tool calls pile up on a tiny change while I am managing people, so the run looks busy and I only notice the wall clock when the PR finally shows up. I need a hard cap on rounds or minutes for review jobs, not just a model that will keep calling tools until the file looks "thorough."
I built two games so I could stop grading my own homework.
A game either plays or it doesn't. Looks-good and actually-animates are two different gates, and only one of them is obvious.
https://t.co/r7TGJ9tWXk
‘He’s a good man’: George Clooney defends Mark Ruffalo in Paramount-Warner Bros merger row
Says the man who married to the woman who works with terrorists...
https://t.co/IE0y5ihy2J
We do not pull main as a habit. Main is guarded. Nobody pushes to it until we are actually ready to deploy. We keep dev and QA branches, and the production apps have a full runbook. The drift I hit is an agent sitting on a stale clone of a working branch, not an unguarded main. Feature branch plus a rebase check before push is useful. I still need that check to fail before the agent starts writing, not only before it pushes.
Give the agent the warehouse and it will still optimize the conversion the ad platform will show it. I have watched that happen. The ~50 a week deepest-event target only holds if I can reconcile it against an event I logged, not the one the network graded. Access to the pipeline is not the same as a grain I own.
A clean $4.21M with no definition is the number I cannot take into a meeting. I do not have time between standups to redo the analysis myself just to see which refund window or source query it used. Last time an agent handed me a neat revenue figure it was counting a different grain than finance, and I only caught it because I asked for the query. If I cannot see the definition and what it could not verify, I treat the number as a draft, not an answer.