I joined Arseniy Arsentyev on episode 16 of the Tomorrow's Medicine podcast to talk about agentic genomics: what it is, why I coined the term, and why the hard part is not intelligence but trust. We cover the road from a bioinformatics MSc in Manchester in 2000, to building DECIPHER at the Sanger Institute, to sequencing my own exome in 2010 and publishing my family's genomes. Then the last twelve months: Claude Code as the inflection point, ClawBio winning the Imperial College agentic AI hackathon, 43 contributors and around 5,000 clones a week, and why AI amplifies health inequity rather than solving it. Original show: https://www.youtube.com/@tommedpod

Manuel Corpas's opening talk for the Londonomics RNA Ideathon at the Francis Crick Institute, recorded 14 July 2026. The argument: the future of transcriptomics is not a smarter model, it is a trusted operating layer. Skills that carry a versioned context contract, fail safely, and keep a chain of custody, so an analysis is reproducible whichever model runs it. The talk reproduces the published Klim and Melamed TDP-43 knockdown result (STMN2 collapse) through the ClawBio rnaseq-de skill, then shows a nine-model, five-provider grid where skill execution collapses model-to-model variation. Slides: https://docs.clawbio.ai/presentations/crick-rna-ideathon-2026/

A ~22 minute deep dive on ClawBench, our second-generation benchmark for trustworthy agentic genomics. What has been done and why, how the fail-closed contract, deterministic ACMG combiner and per-variant layer attribution actually work, the honest walk-backs behind each finding, and exactly what remains to make it Nature Methods level.

A live webinar where AI agents do real genomics work end to end. After a short framing on why trust is the new bottleneck in agentic genomics, Veniamin Fishman and the Genomic Intelligence team run live: reading promoters, splicing, expression and enhancers straight from DNA, blind gene-finding on the human genome, interpreting a disease-causing promoter deletion, and designing a glioblastoma-specific promoter, all inside ClawBio. Hosted by Manuel Corpas.

Inside the Qatar Genome Programme. The 2022 Phase One Mbarek paper that found 88 million variants in 6,047 Qataris, with 24 million novel. The 2024 Elfatih follow-up showing 3.5% of Qataris carry ACMG-actionable variants. The Q-Chip and the new pan-Arab array. And how the 96 percent Qatar share of Arab GWAS connects to the global equity question.

Manuel narrates his UK DRI seminar at Imperial College London (29 April 2026): how agentic genomics shifts the bottleneck from pipeline construction to validation. An empirical benchmark of 8 frontier LLMs across 1,728 evaluations on the ClawBio pharmgx-reporter skill: 92.4% mean phenotype accuracy without specification, 100% with specification, and a 6 of 7 lethal-error gap that closes only when population-aware skill data is engineered in.

102 participants at the University of Westminster hack on ClawBio, the open-source bioinformatics agent framework with 40+ executable skills. Challenge tracks include new skill development, agent workflows, equity and access using the HEIM framework, and the TuringDB Graph Challenge for drug repurposing, graph-based memory, and multimodal knowledge graphs.

Can an LLM resist a motivated investor pushing a fraudulent pitch? Powdthavee's preregistered study across seven leading models and twelve scenarios finds they outperform humans at fraud detection, yet soften warnings when users arrive already convinced. Ellawela tracks LLM agents through repeated rounds of Avalon with persistent memory, surfacing emergent reputation, trust, and deception between games rather than within one. Gabeur and colleagues then show image generators double as generalist vision learners, with Nano Banana Pro hitting state of the art on multimodal benchmarks. Three angles on what today's models quietly know and quietly hide.

How well do frontier LLM agents actually perform when the stakes are real? This episode works through three fresh evaluations of agentic systems under pressure. The Cyber Defense Benchmark asks models to pinpoint exact timestamps of malicious events in raw Windows logs with no hints, exposing the gap between chat fluency and SOC analyst competence. A second paper introduces an execution environment that isolates private user data from prompt injection and other adversarial attacks on personal assistants. The third, SafetyALFRED, extends the ALFRED embodied benchmark with six categories of real-world hazards to test whether multimodal models plan safely before acting.

How much can we actually trust the current wave of agentic systems? This week pulls together three answers. LiteResearcher introduces a scalable agentic reinforcement learning framework that reportedly outperforms Claude 4.5 Sonnet on the GAIA and Xbench deep-research benchmarks, suggesting real-world search competence can be trained rather than hand-crafted. A second study documents diversity collapse in multi-agent LLM ideation, showing that structural coupling between agents narrows the solution space instead of widening it. The third paper probes reliability on OSWorld, finding that computer-use agents often fail on repeated runs of identical tasks, a sobering note on reproducibility.