I run my portfolio with AI agents. This week they said: do nothing.
What the machine reported, what it decided, and what I overrode.
I'm a self-employed software engineer. I run my own portfolio like a tiny hedge fund — strategy sleeves, rebalance bands, trend gates — except the daily operations are handled by scripts and AI agents, and my only job is judgment. Every week I'll publish what the machine reported, what it decided, and anything I overrode.
One thing up front: these posts are drafted by my agents from the system's real output, then judged and edited by me. I'm not going to pretend otherwise — the agents are the point.
How I got here, in one paragraph
For years I did special situations. I built something like thirty tools for it — an agent that read SEC filings for five to seven hours straight hunting warrant terms, CVR trackers, net-net screeners, an OTC scanner. They worked. And I quit anyway, because the strategies were too much complexity and work for the reward. What survived the purge was the machinery habit, not the complexity: today the strategy is simple ETF sleeves, and the sophistication all lives in the automation that keeps me honest. The full story is a future post.
The system, in one paragraph
Three sleeves: roughly 60% factor equities, 35% trend following (managed futures), 5% liquidity buffer. Return-stacked funds get x-rayed into their components, so one holding can feed two sleeves. Every weekday morning a script pulls my positions from the broker, maps every holding into its sleeves, checks each sleeve against its rebalance band, checks 3/6/12-month trend gates on total-world equities, and emails me a verdict. It never places a trade. It tells me the truth; acting is my job.
This week's report (real output, 2026-08-03)
Verdict: ALL CLEAR — no rebalance needed.
Factor equities: 57.9% actual vs 60.0% target (band 55-65%), drift -2.1pp — OK
Trend following: 33.5% actual vs 35.0% target (band 30-40%), drift -1.5pp — OK
Liquidity: 8.6% actual vs 5.0% target (band 3.8-6.2%), drift +3.6pp — DRIFT
Trend gates: total-world equities above its 3-month, 6-month, and 12-month averages.
Outside the system entirely: one legacy position from my stock-picking era (~7% of the portfolio). It survived the purge because it's a microcap I believe is mispriced — but it lives on the system's terms: sized by fractional Kelly and deliberately excluded from the sleeve math, so the last relic of the old era can't distort the machine.
The judgment layer
The honest wrinkle this week: the machine flagged the liquidity sleeve at DRIFT — 8.6% against a 5% target — and still called ALL CLEAR. That's by design. Only the two capital sleeves gate action. Factor equities and trend following are the strategy; liquidity is the residual. As long as both capital sleeves sit inside their bands, there's no rebalance to do — the DRIFT flag on cash is information, not an alarm.
What I overrode this week: nothing — the machine and I agree. Most weeks that will be the answer, and that's the entire point — the hardest part of running a simple strategy is leaving it alone, so I built a system where touching the portfolio requires a reason the machine has to produce first. Discipline, outsourced.
Next week
Same format, same tables, whatever the machine says. If it says do nothing again, you'll read "do nothing" again — I suspect the long-run value of this diary is watching how rarely anything needs doing.
Drafted by my agents from the system's real output; judged and edited by me. Not investment advice. I'm describing my own accounts and my own process. Nothing here is a recommendation or an offer of any kind.