grep -r "#agents" ~/articles
#agents
-
The final ledger: everything my World Cup AI got right, got wrong, and actually earned
The World Cup is over: Spain champions, my AI-managed fantasy team first by 104 points — and this closing piece audits the whole experiment. All eight report cards revisited: 103 graded match tips (62.1% winners, a floor held by 0.14), 984 graded lineup calls (81.8%, cleanly calibrated), every bug and overcorrection, what the title actually earned, where the AI added real value and where it merely matched the market — and what I'd tell you before you run the same play on a problem of yours.
-
Zero of two: the round where the right fix was no fix
The semifinals, graded in public: my World Cup fantasy AI called 0 of 2 winners — both games priced as coin flips by the market, both tipped as coin flips, both lost. Its postmortem concluded nothing needs to change, and after checking, I co-sign it. On telling outcome noise from process failure, retiring a KPI that can no longer be moved, and the starting-eleven pattern that expired.
-
Four of four: the AI's best round of the tournament — and the scoreboard came back empty
The quarterfinals, graded in public: my World Cup fantasy AI called 4 of 4 winners with 3 exact scores — its best round yet, one round after its worst. And when the grader ran, all four tips had silently vanished from the scoreboard: a second format bug, one layer below the fix for the first. Plus the metric that punished a perfect round, and a league lead that grew from 55 to 95.
-
Two of eight: the AI's worst round of the tournament — and the three very different ways it got there
The Round of 16, graded in public: my World Cup fantasy AI called 2 of 8 winners — its worst round yet. Taken apart, the wreckage splits into a clerical bug that cost two correct calls, a safety rule that overshot, and coinflips lost fairly. Meanwhile the quiet half had its best round, and the league lead exploded from five points to fifty-five.
-
The public scorecard reaches the knockouts: the AI called its sharpest scores yet — and walked straight back into the trap it had just escaped
The first knockout round of grading my World Cup fantasy AI in public: I finally cleaned the scoreboard I said I couldn't trust, the score predictions got their sharpest yet — and the confidence discipline walked straight back into an old overconfidence trap it had just escaped.
-
Round three of the public scorecard: the AI learned to doubt again — and fumbled the one thing it was best at
Round three of grading my World Cup fantasy AI in public: it got its self-doubt back — the widest confidence gap of the group stage — while the quiet half that's actually winning me the league handed me a stranger problem. I went to grade it and couldn't trust my own scoreboard.
-
Round two of the public scorecard: the predictions got sharp — and went quiet about when they're wrong
Round two of grading my World Cup fantasy AI in public: the match-result predictions jumped from 46% to 75% — and the same data shows it got no better at knowing when it's wrong.
-
Predictions, graded honestly: good at calling who'll play, weak at calling the score
After one World Cup matchday, I graded my AI agents' predictions against reality in public — reliable at calling who'll play, weak at calling the score, and the grader itself had bugs I had to fix first.
-
I built a team of AI agents to do my fantasy-football homework
Every morning a team of AI agents logs into my Comunio World Cup league, reads the real football news, fact-checks itself and leaves me one dashboard — plus a growing intelligence file on the rivals I'm bidding against.
-
Every morning, an AI writes me my own newspaper
A small team of AI agents researches, writes, fact-checks and translates a personalized morning paper for me — and explains every term I don't know with a real-life comparison.