Guided makes you press Approve on every trade. That's intentional — when you're bringing your own broker connection for the first time, watching every proposal land before any real money moves is the right amount of friction.
After a few months of that, the friction stops being protective and starts being annoying. Most proposals look the same. You approve them. You approve them again. The engine has banked $4,000 in losses for you and you've spent ~30 minutes a week clicking buttons.
That's when Autopilot earns its $99/month.
What Autopilot changes
What does Autopilot tier add to HarvestEngine beyond what Guided provides?
Autopilot adds autonomous execution within configurable safety rails, multi-account household wash-sale pooling, and stop-loss handling — while keeping the full proposal review surface available for optional oversight.
- Auto-execute within rails. Proposals that pass the safety check (caps, kill switch, sector rules) execute automatically.
- Multi-account households. Connect multiple taxable brokerages — Joint, individual, IRA, 529 — and the wash-sale engine pools across all of them.
- Same review surface, optional. You still see every proposal in the dashboard. You can still veto inside the configurable window. Approve becomes opt-out, not opt-in.
- Stop-loss handling. The engine watches positions that approach configured stop levels and proposes (or auto-executes, your call) the harvest.
The autonomy graduation rails
How does the Autopilot autonomy graduation work, and how do users move from shadow mode to full automation?
Autopilot offers five stages — Disabled, Shadow, Small, Standard, and Full — controlled from Settings → Autonomy, allowing investors to start with observe-only Shadow mode before gradually enabling live execution with increasing per-trade and daily value caps.
Autopilot doesn't drop you straight into lights-out execution. There's a five-stage progression you control from Settings → Autonomy:
| Stage | Behavior | Best for |
|---|---|---|
| Disabled | Proposals require manual approve. Same as Guided. | The default state when you first switch to Autopilot. |
| Shadow | The governor logs what it WOULD have auto-approved, but takes no action. You watch the log for a few weeks. | Building trust. Lets you see the governor's judgment without the governor taking action. |
| Small | Auto-execute up to $5K per trade, $25K per day. | Letting the engine handle routine harvests while big trades still bounce to manual. |
| Standard | Auto-execute up to $25K per trade, $100K per day. 30-min veto window on every proposal. | Most users settle here. Real money, real automation, real safety net. |
| Full | No per-trade or daily caps; only the system-wide kill switch. | Multi-million-dollar portfolios where capping at $100K/day leaves harvests on the table. |
The veto window
What is the Autopilot veto window and how does it provide a safety net on auto-approved proposals?
Every auto-approved proposal sits in a configurable veto window — defaulting to approximately 30 minutes — in a ready-to-execute state before it reaches the broker, and investors who reject a proposal within that window cause the engine to cancel it automatically.
Even at full autonomy, every auto-approved proposal sits in a configurable veto window before it goes to broker. Default is 30 minutes. You get a Slack / email / SMS notification the moment the governor approves; the proposal sits in a ready-to-execute state until the window expires; if you click Reject in that window, the engine cancels.
What this gives you: the engine acts on its own confidence, but you can pull the brake within half an hour. Trades the engine made overnight while you were asleep show up as a "here's what I did, here's why" notification when you check in the morning.
The kill switch
What does the Autopilot kill switch do and how is it activated?
The kill switch is a hard system-wide brake in Settings → System that halts all new orders — manual or automatic — immediately upon activation, cancels pending broker orders where possible, and requires explicit confirmation to restore, without destroying any data or proposal history.
Autopilot maintains a hard system-wide kill switch. Anyone with admin access can trip it from Settings → System → Kill Switch. When tripped:
- No new orders, manual or auto, hit any broker.
- Pending broker orders are cancelled where possible.
- The engine continues to compute proposals (so the post-mortem trail is intact) but they are all stamped blocked while the kill switch is active.
Tripping the kill switch never destroys data. Restoring it requires explicit "I understand the consequences" confirmation. It's the panic button.
Time and money saved
How much time does Autopilot typically save compared to manual Guided approvals?
Guided users typically approve 8–15 proposals per month at roughly 2 minutes each; Autopilot at Standard automatically handles approximately 80% of those, reducing active review time from around 25–40 minutes per month to around 5–10 minutes, while also capturing intraday harvests that manual-only users miss.
Concrete numbers from typical users:
- Guided users approve 8-15 proposals per month. At ~2 minutes each, that's 25-40 min/month of clicking.
- Autopilot at Standard auto-handles ~80% of those and routes the remaining 20% to manual review. Time spent drops to ~5-10 min/month.
- Autopilot also catches harvests during the day that Guided users miss because they don't check the dashboard at 10:30 AM. On a $1M portfolio that adds up to ~$2K-$5K/yr of additional captured losses.
When Autopilot doesn't make sense
When does staying on Guided make more sense than upgrading to Autopilot?
Staying on Guided tends to make sense when an investor checks in only weekly (the veto window is less useful then), when the portfolio is heavily concentrated in a single stock, or when the tax year involves unusual realized gains that benefit from reviewing each harvest in the context of the full-year picture.
Many investors choose to stay on Guided when:
- You only check email once a week. The veto window helps when you can react inside it; it's useless if you don't.
- Your portfolio is concentrated (single-stock > 30%). Auto-execution against concentrated positions multiplies risk if the engine misjudges.
- You're in a year where realized gains are unusual (sale of a home, RSU vesting cliff). Manual approve gives you a chance to think about each harvest in the context of the year's total picture.
For everyone else: shadow mode for two weeks, then Standard. That's the path most users take.
Further reading: when Guided makes sense versus staying in Sandbox, what the Alpha tier adds for high-income and concentrated-stock cases, and how harvest pacing through the year affects the autonomy cadence.