AI Drug Discovery Labs

Caribou is the weakest mandate fit in the book — a $1.24 clinical-stage gene-editing micro-cap with no AI or computational platform component, so it belongs to none of the three target buckets. It is also down 3.1% from a $1.29 entry with no catalyst to re-rate it; the Citi $8 target and Moderate Buy consensus are cell-therapy bets, not disruptive-innovation-at-inflection bets. At 3.7% it was too small to matter if it worked and a permanent distraction if it did not. Selling all 323 shares and redeploying into ABCL, which gives the same genomics-innovation exposure through a platform that actually earns partner revenue.

Trimming the book's largest AI-native overweight from 430 to 342 shares — reducing conviction at the margin, not abandoning it. The Recursion OS thesis (phenomics-scale proprietary wet-lab data training its own models) is intact and the position is still +14.5% over a $3.27 entry. But the insider tape has gone one-way: the CEO sold 22,408 shares, a director sold 30,000, all under preset 10b5-1 plans, and the stock fell 3.9% on the session. At 15.2% this was the book's second-largest weight on a pre-revenue platform. At 11.7% it remains the third-largest position and a full-conviction holding, while the freed capital funds both the ABCL opening and the ILMN add — a diversification of the same theme into an engine with partner revenue and an infrastructure name with confirmed momentum.

AbCellera is the bucket-1 name this book was missing: its core asset is an ML-driven antibody discovery engine that partners pay to search, not a single clinical asset, so the platform monetizes across dozens of shots on goal rather than one binary readout. The stock just printed a 52-week high at $12.58 and Phase 2 data is near, with Goldman resuming coverage — an adoption curve at an inflection with a dated catalyst, exactly the setup this mandate is built to own. Opening 43 shares at $12.70 for a ~5% starter, funded by the CRBU exit and the RXRX trim, giving the AI-native platform sleeve a genuinely differentiated second engine alongside RXRX and EXAI.

Illumina is the compute-and-instrument substrate every AI drug discovery platform runs on — no sequencing throughput, no training data. Momentum has turned decisively: a 52-week high, a +8.87% session on top of +7.45% the day before, an index-reshuffle bid, and the position already +24.9% above our $216.18 entry. Adding a 6th share at $270 takes the weight from 12.5% to 14.8% and lifts the computational-biology-infrastructure sleeve (ILMN + NVDA had drifted to 22.6% against a 30% mandate) toward target. This is the one infrastructure holding where the thesis and the tape now agree, so I am concentrating there rather than sprinkling into TWST or TXG.