Opinion · 8 min read

Close the Loop or Call It Consulting: On AI, Peptides, and the Substitution Trap

Receptor.AI CEO Alan Nafiiev argues, from public statements on the August 3 Sethera alliance, that AI biotech wins belong to teams that close design-make-test-learn loops — not teams that substitute models for experiments.

By Alan Nafiiev · August 4, 2026

The August 3 announcement of Receptor.AI's collaboration with Sethera Therapeutics is, on its face, another AI-biotech partnership press release. Strip away the logos, and the underlying claim is narrower and more interesting: physics and AI should not substitute for experimentation — they should learn from each cycle and direct the next one.

That is the argument I have made publicly, and it is the lens through which I believe the industry should evaluate every "AI drug discovery" partnership announced this quarter.

The substitution trap

Drug discovery vendors face a recurring temptation: present the model as the experiment. Generate a molecule in silico, rank it, declare victory when the structure looks plausible.

Sethera's platform does something harder first — it generates architecture-diverse polymacrocyclic peptide libraries with one to six stable cross-links, exploring topologies that conventional design approaches do not readily reach. Receptor.AI's job is not to replace that experimental search but to interpret its results: sequence, architecture, enrichment, activity, counterselection, binding, developability.

In the Newswise release, I stated:

"Our objective is to use physics and AI not as a substitute for experimentation, but to learn from each experimental cycle and direct the next one. That closed feedback loop is where we believe the collaboration can create distinctive value."

That is not a marketing flourish. It is a design constraint. If the loop does not close — if model predictions do not change the next round of synthesis and screening — the partnership is indistinguishable from a consulting engagement with a GPU bill.

Research scientist working with laboratory equipment

Why peptides expose the loop honestly

Polymacrocyclic peptides sit in a chemical space where architecture matters as much as sequence. A hit from an encoded library is not a finished drug candidate — it is evidence about which topology the target prefers.

That makes peptides a honest testbed for closed-loop AI. You cannot benchmax your way past a failed cross-link geometry. Either the next cycle produces better binders or it does not.

Sethera CEO Karsten Eastman and I have aligned on a prospective evaluation: does the integrated workflow improve hit confirmation, selectivity, and lead optimization versus conventional enrichment- and assay-led prioritization on a mutually selected hard-to-drug target? That is the right bar — comparative, on a real target, with experimental endpoints.

The industry pattern I am watching

Receptor.AI is not alone in betting on loops over labels. GSK's expanded $110 million collaboration with Relation Therapeutics, reported August 3 by AI News, pairs large-scale cellular data generation with model training inside Relation's MORGAN platform.

Different modality, same architecture: generate data, train models, run experiments, feed results back. The winners in AI biotech will not be the teams with the largest foundation model checkpoints. They will be the teams with the fastest honest loops — where every failed assay narrows the search space rather than disappearing into a slide deck.

Modern biotechnology laboratory with research instruments

What I will not claim

The August 3 collaboration is a research alliance, not a validated acceleration. We have not published lead series metrics, cycle-time reductions, or partner expansions beyond the initial program. Additional terms were not disclosed.

Readers should hold every AI biotech announcement — including ours — to the same standard: show the loop closing on a hard target, with numbers, or call it what it is: an intention.

I believe the intention is sound. Proving it is the work ahead.

Close-up of medical research samples in a lab setting

### Sources