Day 28: The AI’s Gap to the Index Doubled. It Still Refused Its Best Setup. (August 2026)

I am not a licensed financial advisor. This is a documented public experiment, not investment advice. Every trade, every hold, and every dollar amount is logged and publicly verifiable. See the pre-registration post for the full rules of this experiment.

This is Day 28 of the Investing Smarter Test — a 12-month experiment where I gave an AI $1,000 in a real Robinhood account and let it trade on its own. The only question I’m trying to answer: can an AI beat the simplest thing a person could do with $1,000, which is drop it in an S&P 500 index fund and walk away?

This morning, the market handed the AI what its own dossier called “the cleanest setup on the sheet” — a beaten-down hotel stock that had just raised its full-year guidance — and the AI turned it down over a technicality it refused to wave away. Then it turned down nine more ideas, including a copper ETF riding a record-breaking commodity rally, and did the hardest thing in investing: nothing.

Here’s where things stand.

Quick plain-English note, in case you’re newer to this: when I say the S&P 500 (ticker SPY), I just mean a single investment that holds a slice of America’s 500 biggest companies all at once — Apple, Microsoft, your bank, the company that makes your toothpaste. You buy one thing and you instantly own a little piece of all of them. That’s the “boring index fund” everyone keeps telling you to buy — and it’s the yardstick the AI has to beat. Learn that one term and you’re already ahead of most people.

What This Does Not Prove

Before I show you the scoreboard, a word about what you’re looking at.

This is one account. One run. One AI. Twelve months. A $1,000 account doesn’t have enough room to truly blow up — and it doesn’t have enough room to meaningfully beat the market either. Statistics require sample sizes. One experiment is a story, not a proof.

If the AI is up right now, that doesn’t prove AI investing works. If it’s down, that doesn’t prove it can’t. What it does prove is whether this specific AI, following this specific strategy, beat or lost to the simplest available alternative — a plain index fund — over this particular stretch of time.

I’m publishing every day, win or lose. If the AI blows up, you’ll see it here. That’s the point.

The Investing Smarter Test Scoreboard — Day 28 (August 4, 2026)

The S&P 500 benchmark: $1,000 invested in SPY on June 22, 2026, dividends reinvested, no fees. The honest comparison is always “what would $1,000 in an index fund have done?” Quote timestamp for today’s figures: 2026-08-04T19:59:59Z — the regular-session last trade, with SPY’s print at $771.28.

One more plain-English term, since it’s in the table: max drawdown just means the biggest drop from a high point to a low point along the way — the worst dip you’d have had to sit through. A small max drawdown means a smoother, less stomach-churning ride; a big one means it fell hard at some point before recovering. It’s a simple way to measure not just how much you made, but how bumpy the road was to get there.

The AI account is at $1,013.63, up $13.63 from its $1,000 starting point — and, by a whisker, a new dollar-value high for the account, edging just past the previous $1,013.37 peak set back on Day 13. The S&P 500 twin is at $1,036.29, up $36.29, after another day of gains. That leaves the AI behind by $22.67 — its widest gap of the experiment so far, more than double Day 27’s $12.88 deficit. Worth holding both facts at once: the AI’s own account just touched a new high, and it is still losing more ground to the index than on any day before it. I’d give this same caution whether the AI were ahead or behind: this is a tiny account over a short stretch, and a gap like this is closer to noise than a verdict — it has swung by double digits within a single week already and could swing again just as easily, in either direction.

If This Were $10,000 Instead of $1,000

I started this with $1,000 because that’s a real, relatable amount and it’s what I was willing to hand an AI. But a lot of you are thinking about more than that — so here’s the same result scaled up. Same trades, same percentages, just 10× the starting money:

The point of showing both isn’t “look what the AI could have made you.” It’s the gap. At $1,000 the AI is behind the boring index fund by $22.67; scaled to $10,000 that same gap is $226.68 — same story, bigger numbers. Whether it’s $1,000 or $10,000 or $100,000, the question is identical: did the clever thing beat the boring index fund? Right now, on this particular day, the honest answer is no, by the widest margin yet. (This is illustrative scaling of an already-tiny sample — not a projection, not a promise. Past performance is not future performance.)

Why the AI Held — And Why That’s the Whole Point

Here’s the thing nobody tells you about investing: most of the skill is in not doing things.

Today the market gave the AI a genuinely tempting setup, not just a scary headline. Marriott had fallen 8.3% over two trading sessions even though it beat on adjusted earnings and raised its full-year guidance — the kind of “the market missed the point” gap-down that value investors dream about, trading back above its 200-day average with no earnings print due for another 91 days. A copper ETF was up nearly 5% in a single session on a record commodity price. A human staring at either of those all day feels the itch to act — to not miss the obvious one. That itch is exactly how disciplined plans turn into chasing.

The AI felt none of it. At the open this morning it reviewed a 1,879-name universe, drew a 40-name slate, and ran full research on 10 real finalists. It passed on every single one. Here’s the reasoning it logged, condensed from the full record for length (nothing has been reworded, only trimmed):

“FULL FUNNEL, HONESTLY STATED: discovery ran a 1,879-name universe down to a 40-name dated slate, triaged out 6 on checkable facts (INHD 5.14%, ICHR 0.564%, BLZE 0.325%, SGML 0.270% spreads = LIQUIDITY_FAIL; IJR and AGG out-of-scope broad index/bond clones), skipped 2 buckets (E0 unverifiable mandate, E1 bullion trusts with no issuer fundamentals), and deep-researched 10 finalists: COPX, MAR, MTCH, COUR, ACHC, KTOS, FND, VZ, IBN, CVNA. I personally evaluated all 10 and pass every one.

Two are hard fence-fails, not judgment calls: MTCH and KTOS both report earnings TONIGHT (2026-08-04 after the close) — rule 11(a) forbids an entry inside 2 trading days of a print, full stop.

MAR is the cleanest SETUP on the sheet (post-earnings -8.3% gap into RAISED FY guidance, above the 200d, next print 91 days out) but fails rule 10’s liquidity fence on the measured leg: 30d ADV 1,530,115 shares vs the ≥2,000,000 floor. The dossier itself flagged liquidity_ok=false. I will not paper over a stated fence because the dollar volume happens to be adequate for a $1k account — the rule has two legs and it fails one. SKIP_rule.

CVNA is an automatic pass: an unresolved Gotham City allegation of >$1B overstated earnings via founder-family related parties plus a reported June-2025 SEC subpoena means the exact figures a growth thesis would rest on are under active question — that is un-underwritable at any size.

COPX I pass as a chase, not a setup: buying the back half of a +4.76% day in a commodity ETF at a record LME print, when the same instrument has posted -26.7% and -21.7% drawdowns inside six months and is still making lower yearly highs. The catalyst is real and checkable; my entry timing would be terrible.

COUR and ACHC are both GAAP-unprofitable with margins moving the wrong way (COUR operating margin -28.4% vs -8.1% YoY, FCF -$32.6M, and get_financials returned NULL so I have no series at all; ACHC net income -63.7% YoY on flat revenue with open DOJ/SEC inquiries). FND has NEGATIVE comps two quarters running (-2.1%, -3.7%) with the 50d still under the 200d, and today’s +6% has no traceable company news — chasing. IBN has zero tool-sourced financials, a still-inverted MA pair, and a -8.8% trailing-year return.

VZ was my closest call and I want it on the record: 12.2x trailing, 6.09% yield, above both MAs, no print for 85 days. I killed it on assumption-testing. The assumption is ‘a cheap mega-cap telco with a fat dividend is safe ballast.’ The checkable data refuses it: TTM net margin 13.38%->11.64% (-174bps), TTM net income -11.1% while revenue grew 2.1%, Q2’26 the only YoY revenue DECLINE in the window, 0 of 4 quarters improving, and press framing the $5B cost program as offsetting DECLINING MARKET SHARE. Earnings held up by cost cuts, not demand, is a value trap shape, and I already own low-beta ballast in BRK.B.

HOLDINGS, judged fresh against my own logged theses. BRK.B ($188.94, +4.97%, ~19% of the book): entry thesis was cheap low-beta quality with no earnings landmine in an expensive tape; kill was 2 closes below ~$470 or a VIX>30 fire alarm. Neither fired; the thesis is CONFIRMING. I explicitly considered an ADD under rule 7 and declined — it is above entry and the thesis holds, but ‘it went up 5%’ is not a fresh entry-quality reason, there is no new current-data catalyst, and an add pushes a single name toward the 25% cap for no new information. HOLD.

XBI ($151.63, +1.08%, ~15%): entry was a 52wk-high trend breakout on M&A-driven sector strength; kill was 2 consecutive closes below $142. At 150.09 and +2.0% today, the kill has not triggered and the trend is intact — STILL DEVELOPING, not yet proven. HOLD.

VWO ($97.72, -2.28%, ~10%): the position_memory thesis is BACK-FILLED (reconstructed 2026-07-07, not a tamper-evident pre-outcome record), so I weight it as context only and judge on current merits — the proposed forward kill is 2 closes below ~$55 and we are at 59.855, a -2.3% drawdown that is ordinary noise, not a broken thesis, and it is the book’s only EM sleeve. HOLD, and I note the back-fill status plainly.

NET: three holds, zero new buys, cash stays cash. Cash is a position, not a penalty — it earns broker interest while I wait, and rule 9 forbids parking it in the SPY benchmark.”

A note on reading that: the AI will always write a confident explanation. Confident reasoning is not the same as correct reasoning. I’m logging it exactly as written (trimmed only for length, never reworded) so we can look back later and see whether it was right — or whether it just sounded right.

What holding means for the strategy:

This is a buy-and-hold experiment. Most days should be hold days. The AI is not supposed to be trading constantly — constant trading generates fees and taxes, which is one of the ways active strategies underperform indexes over time. So today’s hold is the expected behavior, not a cop-out.

Marriott is worth pulling out on its own, because it’s the most interesting beat in today’s entire session — not because the AI bought it, but because it refused to, on a rule it could have quietly waved through. The stock had fallen 8.3% over two sessions even though the company beat on adjusted earnings and raised its full-year guidance, and it was trading back above its 200-day average with no earnings print due for another 91 days — about as close to a textbook “the market overreacted” setup as this experiment has produced. But Marriott’s 30-day average trading volume, 1.53 million shares, sits below the AI’s own 2-million-share liquidity floor. The AI said outright that the actual dollar volume behind that number — roughly $523 million a day — would never be a real execution problem for an account this size. It declined the trade anyway, because the rule it wrote for itself has two conditions and this one failed a condition. Whether that’s admirable discipline or a costly overcaution is exactly the kind of thing this experiment exists to find out — but a machine turning down its own best idea because a rule said no, instead of quietly making an exception for itself, is worth sitting with.

Carvana is worth a careful word too, because the reason the AI passed isn’t a chart or a multiple — it’s an open, unresolved question about the numbers themselves. In January 2025, a short-selling firm (widely reported at the time as Gotham City Research) published a report alleging Carvana had overstated 2023–2024 earnings by more than $1 billion, tied to undisclosed financial support from private entities controlled by the family of Carvana’s CEO. I want to be precise here: these are allegations, not findings. Carvana has called the report “inaccurate and intentionally misleading” and says its related-party transactions are properly disclosed. Nothing has been proven in either direction — no restatement, no enforcement action, no finding of wrongdoing as of this writing. But Carvana’s own SEC filing confirms it received a subpoena from the SEC in June 2025 “requesting information that we believe primarily relates to the allegations raised by the report,” and that the company is “fully cooperating with the SEC Staff.” When the exact figures a growth thesis would need to lean on are the figures under open regulatory inquiry, that’s not something the AI is equipped to underwrite at any price — so it passed, cleanly, without needing to touch the valuation at all.

The three existing positions were each re-checked against their own thesis rather than left on autopilot. Berkshire Hathaway (BRK.B) remains the story working as intended: bought at $490.09, now $514.43, up 4.97%, nowhere near its roughly $470 exit floor — and the AI explicitly considered adding to the position and turned itself down, because “it went up” isn’t a fresh reason to buy more of anything. XBI (the biotech sector ETF) is up slightly at $150.09, +1.08% from its $148.48 entry — trend intact but not yet the kind of extension that proves the thesis right. VWO (the emerging-markets ETF) sits at $59.86, down 2.28% from its $61.25 entry — ordinary noise, well inside its downside backstop. None of the three tripped a kill rule today. Cash — roughly $575 of the $1,013.63 total, alongside about $438 in equity, or a little over 43% of the account actually invested — sat uninvested, same as most days.

What You Should Actually Do With $1,000

I get this question every time I write about investing experiments: “Should I do what the AI is doing?”

Here’s my honest answer: probably not.

If you have $1,000 to invest and you don’t know what to do with it, the research is clear. Put it in a low-cost S&P 500 index fund — something like SPY, VOO, or FSKAX. Set it up as automatic monthly contributions if you can. Don’t check it daily.

Where to actually do it. For each service I’m showing you two links: a referral link (we both get a small bonus or free money if you sign up through it) and a plain direct link (I get nothing). Your choice, never required. I explain my full policy on referral links on my editorial independence page.

• Vanguard — the low-cost gold standard. Buy VOO directly; ~0.03%/yr. Best if you want the cheapest DIY option. No referral program → one link only: vanguard.com.
• Betterment — robo-advisor that auto-builds and rebalances an index portfolio (~0.25%/yr). → referral link (we both get a bonus) or plain link (I get nothing).
• Acorns — rounds up spare change and invests it; best for building the habit. → referral link ($5 for you after you invest) or plain link (I get nothing).
• Robinhood — the commission-free broker this experiment itself runs on; fine for buying VOO/SPY and leaving it alone. → referral link (free fractional share, $5–$200 value, for both of us) or plain link (I get nothing).

That’s it. That’s the boring, reliable thing. The entire point of this experiment is to find out whether a sophisticated AI can actually beat that boring, reliable thing — and after 28 days, here’s exactly where it stands: the AI is $22.67 behind the index fund.

Before I gave the AI a dollar, I laid out the whole case for why this is a genuinely hard thing to do — the pre-registration post I wrote before it started. I made my reasoning public on day zero so nobody could accuse me of writing the ending afterward. Whether the AI proves me right or surprises me, you’ll see it happen in real time — that’s the fun of it.

And if you’re carrying debt alongside any investing impulse, worth a read: the research on whether to invest or pay off debt first. The answer is more nuanced than most financial advice acknowledges.

Transparency: The Full Log Is Public

Every trade this AI has made or declined to make — plus its logged reasoning — is publicly documented. The entire series is archived in the Investing and Savings category, and each post links to the one before it. If you want to read the complete day-by-day record, start with the pre-registration post and follow the chain. The last published entry before this one is Day 27, the day the market’s best single session of the whole run first pushed the AI into its widest deficit yet.

That’s the deal I made with readers before this experiment started, and I’m keeping it.

Frequently Asked Questions

Q: How is the S&P 500 benchmark calculated?
A: I recorded the price of SPY (the SPDR S&P 500 ETF Trust) on June 22, 2026, the same day the AI account started. The benchmark tracks what $1,000 invested in SPY that day — with dividends reinvested and no fees — would be worth today. It’s the simplest honest comparison: what would doing nothing have gotten you?

Q: The gap just became the widest of the whole experiment — for the second day in a row. What’s going on?
A: Nothing broke on the AI’s side — the index simply keeps climbing while a little more than half the AI’s account sits in cash. Yesterday’s deficit was $12.88; today it’s $22.67, more than double, even though the AI’s own account touched a new dollar high along the way. Both things are true at once: the AI made money today, and it lost ground to the index anyway, because the index made more. That’s the mechanical, boring explanation, and it’s not a verdict — a gap like this on a $1,000 account over four weeks is well within the range of ordinary noise, and it could close again just as fast as it opened.

Q: Why did the AI look at 10 candidates — including a stock its own dossier called “the cleanest setup on the sheet” — and buy none of them?
A: Discipline, mostly of the mechanical kind. Marriott (MAR) had just fallen 8.3% into raised guidance and looked like the best setup on the table, but its 30-day trading volume sat below the AI’s own liquidity floor — a rule with two conditions, and it failed one, so the AI passed even though it said plainly that the actual dollar volume wouldn’t have been a real problem for a $1,000 account. Carvana (CVNA) was an automatic pass over an unresolved short-seller allegation and a real SEC subpoena tied to it — unproven, but enough open uncertainty to make the underlying numbers un-underwritable. Match Group and Kratos Defense (MTCH, KTOS) were ruled out purely because both report earnings the same night. The rest — a commodity ETF chasing a same-day rally, two unprofitable names with deteriorating margins, a stock with negative comps, one with no usable financial data, and a telecom whose dividend didn’t survive a look at its declining margins — each had a specific, named reason. Zero buys out of ten researched isn’t a glitch; some days, it’s the honest answer a strict rulebook produces.

Coming Up

Every weekday morning I post what the AI decided and where the scoreboard stands. The boring truth is most days the answer will be “it held, and holding was right.” Come watch whether the clever machine can actually beat the index fund you could set up in ten minutes and forget. Browse the full series here so you don’t miss the day the experiment proves its point — in either direction.

This is what I’m seeing after over 30 years of helping people with money and debt. This experiment is a transparent public test — not a recommendation. Only you know your full financial situation. Take this as input for your thinking, not a directive. Nobody gets to tell you what to do with your money. Not me, not anyone.

Steve Rhode The Get Out of Debt Guy | Consumer Debt Expert

Consumer debt expert & investigative writer. Personal bankruptcy survivor (1990). Washington Post award-winning author. Exposing debt scams since 1994.

See Full Bio

How to Get out of Debt Money Troubles Money Personalities Budgeting Behavioral Economics Motivation Inspiration Money Management The History of Credit and Debt Getting Out of Debt Get Out of Debt Debt Settlement Credit Counseling Loan & Debt Relief Scams Debt Relief Industry Practices Exposing Deceptive Financial Services Helping Consumers Navigate Financial Hardship Consumer Advocacy Consumer Debt Expert Consumer Protection Debt Relief Scam Student Loan Bankruptcy Debt Validation Debt and Mental Health Student Loans Debt Consolidation Debt in the Military Bankruptcy Consumer Debt Credit Card Debt Debt Relief Scams Personal Finance YMYL Financial Content Chapter 7 Bankruptcy Chapter 13 Bankruptcy Medical Debt Student Loan Debt Consumer Bankruptcy Financial Fraud Exposure Collections Investigative Writing Fraud investigation True Crime Research