One notable place this happens is backtesting. Particularly for passive strategies, you don't actually know from historical data what would have happened if you had been in the market at a given time. Would your prices move the market? Maybe. Would you get filled disproportionately when it's bad to get filled? Fair chance of it. So don't act like your presence would not be felt.
Another angle is risk management. You may have high confidence in your model, but you should have a plan for what to do if things don't go as expected. It seems crazy to me that it could get so far out of hand that a quarter of the company gets sacked. Why not set a budget and dip your toes? Then you can watch out for the adverse selection issues as you grow the portfolio and make adjustments.
Finally, it's very normal for quant strategies to look good when they're first proposed, then look a bit worse when traded in real life, and perhaps then degrade over time. Another reason to take it easy when sizing the bets.
You're right about Lucas critique. It's fundamental that models affect the thing they're modelling.