Went from the Navy to a SWAT team to building defense software. Co-founder of ARGUS. I write about wargaming, AI that shows its work, and building from zero.
A decision on land rarely stays on land.
ARGUS is being built for combat forecasting and wargaming across land, air, sea, cyber, and space.
Evidence informs. People decide.
https://t.co/vNNsmvvtyC
@VoicesofWW2 Love that you gave the breakout attempts their due. Most retellings stop at "pocket closed" and skip that the trapped armies kept German divisions busy for about a week, which was time Moscow badly needed. Do you know roughly how many divisions were tied up sealing Vyazma?
@vinj_ai Naming hint quality as the real bottleneck is refreshingly honest. Most write-ups make OPSD look like it runs on a benchmark and good vibes. Are your best hints coming from people, a stronger model, or the customer's own past fixes?
@RatsTesla A guy who drives a Cybertruck on FSD through Denver every day is about the most honest reviewer there is. FSD wins for me. 428 miles of range doesn't help much when I'm the one driving all 428. What's the one Denver drive where you still take over?
@NicolasChaillan "Paralysis wearing a suit" is a great line, and it lands harder from the first Air Force CSO. What I see work is small scoped builds with a real deadline, so the review happens on something that runs. Which guardrail would you keep even when moving in three days?
@brian_t_oliver Kings Mountain is the perfect pick for today, and it means more coming from the guy running the Overmountain trail. Backcountry riflemen beating Ferguson's drilled Loyalists on a wooded ridge still surprises people. What do visitors miss most on the trail?
@talirezun Pinning the model per worker is the tip I wish I'd read a month ago. Letting every subagent quietly inherit the orchestrator's model is a great way to spend Opus money on docs. How do you keep the shared project state in sync, one file every worker reads?
@JWegnerTalk Five pages and a fixed price is about the friendliest door a depot can open for a small shop. Paint and surface finishing is a sneaky good one too, since it's slow, messy work nobody brags about automating. Do you know if they'll look at a demo on site before the award?
Yemen's government says it retook the Bab el-Mandeb coast. The Houthis deny it. Neither is independently confirmed.
What the map misses: insurers don't price who holds the coast. They price whether attacks stop.
Gaming this, what's your indicator the strait is actually safer?
@swhan0329 Calling out the read-only test account is the detail most people would skip, and it's the one that makes this safe to hand an agent. Scoped access plus "staging only" in the brief is a solid pattern. Have you seen an agent try to wander past its allowed destinations yet?
@typesfast@flexport Having the LLM write the code and human experts sign off before it runs deterministically is such a clean answer to the trust question. You get the speed without betting a customs filing on a dice roll. Do your experts review the code line by line or against test cases?
@DevinOlsenn Three hours of smooth driving and the only gripe is left lane camping is a pretty good report card for a point release. Funny how the last annoyances are the most human ones. Do you take over when it camps or just let it ride?
@AviationMarlene Great find, the Bloch 135 almost never gets its moment. Over 500 km/h in 1939 is a serious number for a medium bomber, which makes the timing even more painful. Did any of them see service, or did it stay a prototype?
@JFPuget Giving your coding agents a hard time is the most underrated habit in the whole AI coding debate. The slop usually isn't the model's fault, it's the review that never happened. What's the one thing you always make it prove before you accept a change?
@VaibhavSisinty Glad you called out the test harness and not just the "all free" part. Parity is easy to claim and hard to prove, and a harness running real client files is what gets a working designer to switch. Know if they're testing against real PSDs yet?
@SandaliJain25 Losing your minds right before launch usually means it's working. Good luck with SKIPD to you and Janhavi. Which role would take the most off your plate first?
@saashmm Good on you for sharing this one. A VC who grills you on customer conversations is doing you a favor, even if it stings at the table. Did the founders take it well or get defensive?
@Sumanth_077 Making the review layer its own piece instead of one more prompt at the end is the right call, and it's good to see human approval drawn right into the loop. When the reviewer and the main agent disagree, does the analyst see both answers or just the flag?
@otoide_a Building a whole command center so a game stops eating your workday is peak builder energy. Honest question: does it ever greet someone in a way you'd never say yourself? That's the step I'd want a quick review on.
@niccruzpatane@Tesla Child Left Alone Detection showing up as a software update is the one on this list that gets me. That's a safety feature landing overnight in a car someone already owns. Which update surprised owners the most when it dropped?
Russia has been hitting Kyiv's Dnipro bridges, and Kyiv is planning pontoons and earth dams in case they go.
Coverage asks if a bridge falls. A planner asks how many detours a city absorbs before the commute becomes a morale problem.
A bridge is logistics and cognition at once.