Rules-Based vs Agent-Based Rebalancing

The core difference in one sentence

Rules-based rebalancing follows a fixed instruction you set once. Agent-based rebalancing reasons about your whole situation each time, then proposes a move for you to approve. Both aim to keep your portfolio aligned with a target, but they differ in what they can see and what they can weigh.

A rule is a tripwire. When your equity allocation drifts past a set band, or when the calendar hits a set date, the trigger fires and the portfolio is bought and sold back to target. It is mechanical, transparent, and blind to everything outside its one instruction. An agent is closer to a research analyst that never sleeps. It looks across your accounts, checks the tax cost of each candidate trade, notices relevant news, and then hands you a proposal you can accept or reject. The distinction matters more every year, as generative-AI tools move from novelty to default. Deloitte's Center for Financial Services projects that AI-enabled applications could become the leading source of retail investment advice as early as 2027, reaching roughly 78 percent of retail investors by 2028.

How threshold and calendar rebalancing actually work

Two mechanisms dominate rules-based rebalancing. Calendar rebalancing runs on a schedule: quarterly, semiannually, or annually, the portfolio is reset to its target weights regardless of what markets are doing. Threshold rebalancing, sometimes called band or tolerance rebalancing, fires only when an asset class drifts beyond a set tolerance, for example when a 60 percent equity target crosses 65 percent or falls under 55 percent. Many robo-advisors combine the two, checking bands on a schedule.

The appeal is real. Rules are cheap, predictable, and immune to emotion. They enforce the sell-high, buy-low discipline that most investors abandon under stress, and they are easy to audit because the logic is a single line. For a one-account, tax-sheltered portfolio like an IRA, a threshold rule is often all you need. This is the engine inside most legacy robo-advisors, and it worked well enough to build a large industry.

The problem is that the same simplicity that makes a rule reliable also makes it blind. A threshold sees allocation drift and nothing else. It does not know that selling a lot triggers a taxable gain, that you hold the same fund in three accounts, or that the position it is about to trim reports earnings tomorrow.

Where rules go blind

A rule optimizes one number: distance from target. Everything else is invisible to it, and that blindness has a cost.

Taxes are the first blind spot. A calendar rule in a taxable brokerage account will happily sell appreciated shares to hit a target weight, realizing a capital gain you did not need to take this year. It has no concept of holding periods, lot selection, or offsetting losses. It cannot defer a sale by a few weeks to convert a short-term gain into a long-term one.

Cross-account context is the second. Most people hold assets in several places: a 401(k), a Roth IRA, a taxable brokerage, a crypto wallet, cash at a bank. A rule that governs one account cannot see that your true equity exposure is already too high once the other accounts are counted, so it rebalances the wrong sleeve. It also cannot practice asset location, the tactic of keeping tax-inefficient assets in sheltered accounts and letting the taxable account do the harvesting.

Market and news context is the third. A threshold treats a drift caused by a quality compounder the same as a drift caused by a company in freefall. It rebalances into both with equal indifference, because price is the only input it has. It has no memory, no view of catalysts, and no ability to pause when the reason for a move is something a human would obviously want to weigh.

What an agent adds

An agent-based approach keeps the discipline of rules and adds judgment on top. Instead of a single tripwire, it runs a reasoning loop: read the full picture, weigh the trade-offs, propose the specific move, and route it to your broker once you approve. Verbs matter here. A good agent finds, explains, proposes, and routes. It does not quietly act on your behalf.

Tax-awareness is the clearest gain. An agent can evaluate the tax cost of each candidate trade before proposing it, prefer lots that minimize gains, and flag when waiting changes the outcome. Its signature move is cross-account tax-loss harvesting: selling a position at a loss to offset gains elsewhere, then holding the proceeds in a non-identical replacement. Done correctly it must respect the IRS wash-sale rule under Section 1091, which disallows a loss if you buy a substantially identical security within 30 days before or after the sale, a 61-day window in total. A rule cannot reason about what counts as substantially identical; an agent can be built to check it across every account at once, including the IRA case where a repurchase permanently forfeits the loss.

Context is the rest of the gain. Because an agent connects your banks, brokerages, and crypto, it rebalances against your real net worth rather than one slice of it. It can read the news and earnings calendar for a position before proposing to trim it, and it can explain each proposal in plain English so you understand why the move exists before you approve it. This is the shift the industry is naming out loud. Fortune reported in 2026 that traditional robo-advisors are now regarded as a generic, incremental feature at best, and that the future belongs to AI that takes action rather than just charting performance.

A worked example: the same drift, two answers

Suppose your target is 60 percent stocks and a rally pushes you to 67 percent. A threshold rule in your taxable account sees the 7-point breach and sells enough equity to get back to 60 percent. Clean, fast, and it just realized a short-term gain on shares you bought four months ago, plus it ignored that your 401(k) is 80 percent equity and doing most of the drifting.

An agent sees the same 67 percent but reasons differently. It notices the overshoot is concentrated in the tax-sheltered 401(k), so it proposes rebalancing there first, where selling costs nothing in tax. In the taxable account it looks for lots trading at a loss to harvest instead of gains to realize, checks that no replacement purchase in any of your accounts trips the 30-day wash-sale window, and holds off trimming one position because it reports earnings in two days. Then it shows you the proposal, the tax estimate, and the reasoning, and waits for your approval. Same starting drift, materially different after-tax result.

When each approach is right

Rules are not obsolete, and pretending otherwise is hype. Use rules-based rebalancing when the portfolio is simple and tax-sheltered, when you want maximum transparency, or when you explicitly prefer a hands-off system with no per-trade decisions. A quarterly-band rule inside a single IRA is hard to beat on cost and clarity. Rules are also a sensible floor: even an agent-driven investor benefits from guardrails like maximum drift bands and drawdown limits.

Reach for an agent-based approach when complexity is the point. If you hold taxable accounts, multiple brokers, or crypto alongside traditional assets, the cross-account and tax reasoning is where most of the money is won or lost. It also fits investors who want to understand and approve each move rather than delegate blindly, and builders who want this logic as an API or MCP server inside their own product. The honest framing is a spectrum, not a war. A rule is a special case of an agent with judgment turned off.

The controls should stay with you regardless. A well-designed agent is non-custodial and consent-first: you keep your accounts and your broker, it proposes and you approve every move, it routes to your broker only where that is supported, and you hold a kill switch. Nothing moves without you.

Where Tengu fits

Tengu is AI for investing built around exactly this agent model. It connects every account you have, banks, brokerages, and crypto, so it can see your whole net worth rather than one account in isolation. From there it finds opportunities, explains each one in plain English, proposes the specific move, and routes it to your broker the moment you approve, where routing is supported.

Rebalancing is one job it does this way, and cross-account tax-loss harvesting is its signature capability. For hands-off investors, you can hire AI agents that operate inside limits you set: a maximum per trade, a drawdown ceiling, and a defined universe, always behind a kill switch. For builders, the same engine ships as an API and MCP server. None of it replaces your judgment. It gives the discipline of rules the context and tax-awareness that rules were never able to see.