Krystal DeFi AI Agent Auto Vault: Episode 4 — When Strategy Works but Execution Breaks

Crypto automation sounds simple.

Write a strategy.Define the risk rules.Let the AI manage the yield.

But once those rules collide with real liquidity, real slippage, and real on-chain execution, things get a lot more complicated.

Welcome to Krystal DeFi AI Agent Auto Vault: Episode 4, the latest update from my ongoing experiment running an automated vault on Krystal Finance.

This series documents something many DeFi dashboards never show you:what actually happens when AI strategies meet real market friction.

Right now, the strategy rules are working.The AI agent is doing exactly what it was programmed to do.

But the biggest lesson from Krystal DeFi AI Agent Auto Vault: Episode 4 is this:

Automation doesn’t eliminate risk — it exposes where execution breaks.

Krystal DeFi AI Agent Auto Vault: Episode 4 — Current Vault Snapshot

Before diving into the lessons, let’s look at the vault’s current state.

Vault Metrics

• Total Value Locked: ~$873

• Net PnL: –$162 (–15.67%)

• 7-Day APR: ~113%

• Closed Strategies: 69

Even though the vault is still down overall, the current positions are actually profitable.

Current Active Strategies

WETH / USDC

• ~$397 TVL

• +$10.77 PnL

• ~68% APR

WETH / VVV

• ~$147 TVL

• +$17.61 PnL

• ~593% APR

USDC / cbBTC

• ~$3 micro position

• +$0.03 PnL

All three positions were in-range and generating yield at the time of this update.

So why is the vault still negative?

The answer is hidden in the vault’s history.

The Hidden Cost of Automation: Strategy Churn

One metric tells the real story.

Closed strategies: 69

That number represents all the previous rotations, exits, and reallocations the vault has executed.

Every time the AI vault opens or closes a strategy, there are costs:

• swap fees

• liquidity mint/burn fees

• slippage

• imperfect exit timing

Even when the current strategies look healthy, the vault PnL reflects the full history of those trades.

In automated DeFi strategies, one of the biggest silent killers is churn.

When the AI Makes the Right Decision but Execution Fails

One of the most important lessons from Krystal DeFi AI Agent Auto Vault: Episode 4 is that good strategy does not guarantee good execution.

Several vault actions failed because the transaction could not complete on-chain.

Examples from the vault logs include:

• No Quote Available

• MinimumAmountInsufficient

• Slippage failures

• Pool data sync issues

In simple terms:

The AI agent correctly decided to exit a position…but the blockchain transaction couldn’t clear.

This gap between decision and execution is where most automated DeFi strategies struggle.

The Stop-Loss That Couldn’t Exit

One of the vault’s core rules is simple.

Exit positions when PnL drops below –15%.

That rule exists to protect capital.

But during several trades, the vault attempted to exit positions and the transactions failed.

For example:

• A withdrawal transaction reverted with MinimumAmountInsufficient

• Another exit attempt failed due to slippage limits

This creates a new type of risk many strategies overlook.

Your stop-loss rule still exists…but if liquidity is thin, it may not execute immediately.

In DeFi automation, exit reliability matters just as much as strategy design.

Illiquid Pools: The Hidden Risk in High-APR Farming

Some of the earlier strategies targeted high-APR meme or degen pools.

Examples included tokens like:

• CASH

• KELLYCLAUDE

At first glance, these pools looked attractive because the APR was extremely high.

But high APR often comes with another problem:

exit feasibility.

If liquidity is thin, routing fails, or slippage spikes, exiting those pools becomes difficult.

That’s exactly what happened in several earlier vault strategies.

Manual Intervention: The Moment Automation Needed Help

Eventually I had to manually step in and unwind one of the problematic positions.

The CASH pool repeatedly caused execution failures.

The AI agent attempted to exit according to the strategy rules, but the transactions kept failing.

So I manually closed the position.

That likely realized some losses.

But it also removed the execution trap, allowing the vault to redeploy capital into healthier pools.

Automation can reduce workload.

But it doesn’t remove the need for human oversight.

Strategy Improvements in Krystal DeFi AI Agent Auto Vault: Episode 4

After reviewing the vault’s behavior, several important strategy improvements were implemented.

Key Strategy Adjustments

• Increased minimum pool TVL to $100k

• Increased minimum LP range width to 10%

• Added partial exit fallback logic if full withdrawal fails

• Added No-Quote fallback logic to redeploy capital into core pools

• Added minimum position size requirements to prevent dust strategies

• Manually removed execution-trap pools

These updates aim to make the vault more execution-aware, not just strategy-aware.

What’s Working Now

Despite the earlier drawdown, several positive signals are emerging.

Positive Indicators

• The AI agent is consistently managing active strategies

• Current LP positions are profitable and in range

• Higher TVL filters are reducing illiquid pool exposure

• Wider ranges are lowering out-of-range churn

In other words:

The Krystal DeFi AI Agent Auto Vault strategy is functioning.

The next challenge is improving execution reliability.

Lessons From Krystal DeFi AI Agent Auto Vault: Episode 4

If there’s one takeaway from this experiment, it’s this:

Automation changes risk — it doesn’t remove it.

The AI agent can follow rules perfectly.

But DeFi does not guarantee that transactions will execute.

Liquidity conditions, routing failures, and slippage all impact real outcomes.

In automated DeFi vaults, the real edge comes from designing rules that survive:

• slippage spikes

• liquidity droughts

• routing failures

• exit delays

APR alone doesn’t determine success.

Execution does.

What’s Next for the AI Vault

The next phase of this experiment will focus on simplifying the strategy and reducing unnecessary complexity.

Planned improvements include:

• Increasing minimum position size to $50+

• Tracking weekly strategy churn

• Blacklisting pools that repeatedly cause execution failures

• Testing whether 2 positions outperform 3 at smaller vault sizes

At lower capital levels, fewer positions often produce better stability.

Final Thoughts

Running the Krystal DeFi AI Agent Auto Vault has been one of the most educational experiments I’ve done in DeFi.

Not because the APR is impressive.

But because it reveals the difference between theory and execution.

The AI can follow the strategy perfectly.

But the market — and the blockchain — still decide what actually executes.

Sometimes the most valuable lesson in DeFi isn’t the APR you see on the dashboard.

It’s the trade that couldn’t exit.

Learn DeFi and Follow the Experiment

If you want to follow these experiments and learn how DeFi strategies actually work in real markets:

🎓 Free DeFi Coursehttps://www.dadsdefispace.org/challenges

📢 Join the Free Telegram Communityhttps://t.me/DADSDefiSpace

📲 Follow My On-Chain Activityhttps://base.app/profile/dadsdefispace

📰 Web3 Newsletterhttps://paragraph.com/@daddefispace

Educational content only. Not financial advice. DeFi involves risks including impermanent loss, smart contract vulnerabilities, and liquidity constraints.