7 min read

Agents reach for the treatment plan

Agents reach for the treatment plan
Nº 01 · The Lede arXiv Agents · Infrastructure

Agent drafts colorectal treatment plans

Agent drafts colorectal treatment plans
Fig. IarXiv · Filed 11 Aug 2026.

Agentic LLM plans colorectal cancer treatment in a new arXiv paper, applying a generative model that works through a case in steps rather than answering in one shot. Oncology planning has been the hard case for clinical language models, since it depends on staging, prior therapy, comorbidity and guideline branching that a single prompt rarely captures. Putting that into an agent loop moves multi-step clinical reasoning into the same category as the retrieval and summarization tasks models already handle competently. Prospective evaluation against tumor-board consensus remains the bar this class of system has to clear, and a preprint doesn't clear it.

Read the source

Medical AI bias persists
Fig. IIHacker News · Filed 11 Aug 2026.
Nº 02 Hacker News Field report

Medical AI bias persists

Medical AI bias persists in newer models, Flinders University researchers report, finding that current systems still reproduce racial and gender stereotypes in medical contexts. The assumption that each model generation quietly sands down the bias of the last one doesn't survive the check. That keeps bias auditing a standing line item in clinical AI evaluation rather than a solved problem inherited from model vintage, and it raises the evidentiary burden on anyone claiming a system is ready for patient-facing use.

Read more
Phenotypes steer molecule generation
Fig. IIIbioRxiv · Filed 11 Aug 2026.
Nº 03 bioRxiv Field report

Phenotypes steer molecule generation

Generative design conditioned on cell phenotypes shows up in a new bioRxiv preprint that joins transcriptomic profiles with morphological readouts into a single conditioning signal for molecule generation. Phenotype-driven design has mostly leaned on one modality at a time. Combining them lets a generator aim at the cellular state a compound produces rather than at a named protein target, which extends generative chemistry into disease biology where the target is unknown or contested.

Read more
Also Filed · Four Briefs from the queue
Nº 04 arXiv Benchmarks · Evaluation

Benchmark for conversational triage

ELICITED tests multi-turn clinical triage with an arXiv benchmark of EHR-grounded longitudinal conversations, where models must seek missing information before committing to a decision. Shifts clinical evaluation from one-shot answer accuracy toward whether a system recognizes what it still needs to ask.

Read
Nº 05 bioRxiv Computational biology

Pharmacogenomics gets a yardstick

Pharmacogenomic variant interpretation benchmarked in a bioRxiv assessment of computational models for drug-response variants. Pathogenicity prediction has had shared yardsticks for years; pharmacogenomics has not, leaving precision-dosing tools with no common ground for comparing accuracy claims.

Read
Nº 06 X Field report

Structure prediction, four times faster

Protein structure prediction runs 4x faster on Nebius AI Cloud using NVIDIA NIM inference containers than launching containers natively, per Nebius, in work with Seqera Labs. Structure prediction is commodity enough now that the competition has moved from model quality to container startup overhead.

Read
Nº 07 X Field report

Anatomy of folding models

A thread on folding-model anatomy argues that AlphaFold- and ESMFold-class systems decompose into two pieces: a sequence encoder, and a trunk plus structure module. The framing points at where components could be swapped to trade accuracy against speed in structure prediction.

Read

Reply with your discoveries. A human reads them. Forward freely.

Agentic Discovery  ·  Nº 74  ·  11 Aug 2026

Editor's Note

Today the agents move past summarizing charts and start proposing what to do about them.

 

Nº 01 · The Lede  —  arXiv  —  Agents · Infrastructure

Agent drafts colorectal treatment plans

Agent drafts colorectal treatment plans

Fig. I  arXiv · Filed 11 Aug 2026.

Agentic LLM plans colorectal cancer treatment in a new arXiv paper, applying a generative model that works through a case in steps rather than answering in one shot. Oncology planning has been the hard case for clinical language models, since it depends on staging, prior therapy, comorbidity and guideline branching that a single prompt rarely captures. Putting that into an agent loop moves multi-step clinical reasoning into the same category as the retrieval and summarization tasks models already handle competently. Prospective evaluation against tumor-board consensus remains the bar this class of system has to clear, and a preprint doesn't clear it.

Read the source →

Why it matters

Treatment planning is where agentic decomposition, not longer context windows, is the claimed unlock — and it shifts the clinical AI debate from whether a model can read the chart to who signs off on the plan.

 

Nº 02  —  Hacker News  —  Field report

Medical AI bias persists

Fig. II  Hacker News · Filed 11 Aug 2026.

Medical AI bias persists

Medical AI bias persists in newer models, Flinders University researchers report, finding that current systems still reproduce racial and gender stereotypes in medical contexts. The assumption that each model generation quietly sands down the bias of the last one doesn't survive the check. That keeps bias auditing a standing line item in clinical AI evaluation rather than a solved problem inherited from model vintage, and it raises the evidentiary burden on anyone claiming a system is ready for patient-facing use.

Read more →

The Bench NoteFrom Heureka Labs

The field is settling into a habit worth having: measure a system's behaviour in the setting where it runs, and keep measuring.

Repeatable checks need artefacts. Run Analysis writes the figures, tables and the script that produced them into the project, so a check can be re-run by someone else instead of taken on summary.
Looked up, with the source attached. ARC's research tools reach literature and reference databases to resolve a claim rather than recall it, and the citation lands in the record beside it.
Lineage captured as work happens. Provenance records how a result connects back to the sample, dataset or experiment it came from, at the time rather than in reconstruction.

What we’re watching: whether per-deployment audits start being published alongside the systems they cover, the way benchmark scores already are

 

Nº 03  —  bioRxiv  —  Field report

Phenotypes steer molecule generation

Fig. III  bioRxiv · Filed 11 Aug 2026.

Phenotypes steer molecule generation

Generative design conditioned on cell phenotypes shows up in a new bioRxiv preprint that joins transcriptomic profiles with morphological readouts into a single conditioning signal for molecule generation. Phenotype-driven design has mostly leaned on one modality at a time. Combining them lets a generator aim at the cellular state a compound produces rather than at a named protein target, which extends generative chemistry into disease biology where the target is unknown or contested.

Read more →

 

Also Filed  ·  Four Briefs from the queue

Nº 04  —  arXiv  —  Benchmarks · Evaluation

Benchmark for conversational triage

ELICITED tests multi-turn clinical triage with an arXiv benchmark of EHR-grounded longitudinal conversations, where models must seek missing information before committing to a decision. Shifts clinical evaluation from one-shot answer accuracy toward whether a system recognizes what it still needs to ask.

Read →

Nº 05  —  bioRxiv  —  Computational biology

Pharmacogenomics gets a yardstick

Pharmacogenomic variant interpretation benchmarked in a bioRxiv assessment of computational models for drug-response variants. Pathogenicity prediction has had shared yardsticks for years; pharmacogenomics has not, leaving precision-dosing tools with no common ground for comparing accuracy claims.

Read →

Nº 06  —  X  —  Field report

Structure prediction, four times faster

Protein structure prediction runs 4x faster on Nebius AI Cloud using NVIDIA NIM inference containers than launching containers natively, per Nebius, in work with Seqera Labs. Structure prediction is commodity enough now that the competition has moved from model quality to container startup overhead.

Read →

Nº 07  —  X  —  Field report

Anatomy of folding models

A thread on folding-model anatomy argues that AlphaFold- and ESMFold-class systems decompose into two pieces: a sequence encoder, and a trunk plus structure module. The framing points at where components could be swapped to trade accuracy against speed in structure prediction.

Read →

 

· · ·

Reply with your discoveries. A human reads them. Forward freely.