We read the reviews of Composer's 'Trade with AI'. Here is what holds up

Composer sells the shortest path from idea to automated strategy: type a rule, let the AI turn it into a 'symphony', backtest it, then let the platform execute it in a brokerage account. SoFi completed its acquisition of Composer Securities in April 2026 and relaunched the product as Composer by SoFi on June 23, 2026, according to a September 23 review by AgentTrading that cites SoFi's press release. We read what the reviews found. Plan details are tracked on our Composer tool page.

What the AI part actually does

Composer's help center describes 'Trade with AI' as the Discover tab of the app: it surfaces existing community strategies or drafts new ones from plain English, each with its logic, backtest and risk stats. Two rules are explicit: you review and confirm every symphony before it runs, and usage is metered by a weekly token allowance that resets every 7 days and depends on your plan. Composer does not publish the token amounts.

What it costs (read October 14, 2026)

Source: Composer pricing. Nodes are the blocks in a strategy's decision tree, so complex rules can push you up a tier even with one strategy. Regulatory fees on sales apply on top. Composer Securities LLC is the broker-dealer, a FINRA and SIPC member per the same page.

Review by review

Alpha GainDaily, hands-on, February 24, 2026. The reviewer tried about six prompts in the AI strategy generator; roughly four produced workable starting frameworks and two produced logic that was valid but strategically poor, such as rebalancing conditions that fired almost every day. The same review reports live results lagging backtests by 'roughly 8-15 percentage points annually' after 30 days on three strategies. That figure rests on one month and three strategies, with no interval: treat it as an anecdote, not a rate nContext: n=3 strategies · 30 days, early 2026 · Alpha GainDaily review · margin: not disclosed.

Alpha GainDaily, second review, June 2026. A later review on the same site calls the AI suggestion layer 'thin', producing 'plausible-looking strategies that were essentially repackaged versions of standard momentum frameworks.' It reports a dual-momentum strategy underperforming its backtest by about 3 to 4 points annualized over six weeks nContext: n=1 strategy · six weeks, 2026 · Alpha GainDaily review · margin: not disclosed, and one rebalance filling about 3 minutes late across 60-plus scheduled events.

AgentTrading, September 23, 2026. The most detailed review reads Composer's own documentation and finds three things worth knowing:

• Backtests use daily adjusted closing prices back to 1990, model regulatory fees, and apply slippage of 1 basis point by default. The reviewer calls that optimistic for leveraged ETFs or trades near the close and suggests re-running at 5 and 10 basis points.
• Pro, and only Pro, lists 'priority automated trade timing closer to market close.' Since backtests assume the close, the cheaper plans may see a gap the backtest never shows, most of all on daily rotation strategies.
• Composer's help center still models a '$40/month' subscription in backtest costs, a plan that no longer exists on the pricing page.

Trustpilot and app stores. The June review notes that Composer's Trustpilot page had fewer than 50 reviews when it checked, too few to read literally. AgentTrading reports Trustpilot complaints about annual fees users did not expect; Pro is shown as a monthly figure but billed yearly.

Where the reviews contradict each other

The February review says backtest data goes back 'roughly a decade for most assets' and describes a '$30/month Maestro plan'; the September review, reading current documentation, says 1990 and lists no such plan. Pricing and plan names reset after the SoFi relaunch, so any review written before mid-2026 describes a different product. Check the date before you trust a review's numbers, including this one.

What the marketing shows

Community strategy pages on Composer display backtested annualized returns and Sharpe ratios next to an 'Invest' button. We do not repeat those figures. Composer's own documentation, quoted by AgentTrading, says 'all backtest results are hypothetical and benefit from hindsight', and the June review warns that some shared strategies are 'backtested junk optimized to look good on historical data.'

Using it with open eyes

• Start on Starter: one automated strategy and unlimited backtests cost nothing.
• Raise slippage to 5 or 10 basis points and re-run before trusting any curve.
• Treat AI-drafted symphonies as a first draft. Two of six in one test were valid code with bad trading logic.
• Browsing 2,000-plus community curves until one looks good is a search, not research.