Steering Phage Therapy

01What changed

• AI-designed bacteriophage reached a first patient treatment with FDA clearance in 2026, moving phage from compassionate-use anecdote to an engineerable therapeutic.1AI-designed phage, first patient, 2026
• Expanding, polymicrobial phage libraries have also cleared FDA to be tested, which means the bottleneck is shifting from finding phages to choosing among them.2Phage directory, 2026

02The gap

Phage selection optimises for lysis in a plate assay. Almost nobody measures what the surviving bacteria become. Resistance is treated as a failure of the phage rather than as a design variable of the cocktail.

03The idea

Design for the escape, not the kill. Choose phage pairs whose receptor requirements are antagonistic: adapting to one receptor imposes a measurable cost on the other. Add a third element, a CRISPR or toxin payload delivered by a non-replicating phage, that converts resistance into a metabolic penalty.

The screening instrument is an evolution arena: continuous culture with a two-phage mixture and a competitor strain, run for thirty days with sequencing every twelve hours. The readout is not minimum inhibitory concentration but the trajectory of resistance alleles and the growth rate of the dominant mutant.

Rank candidates by a steering index: the fitness cost of the most accessible escape mutation divided by the time it takes to appear. A high index means resistance exists but is expensive, which is the regime where a therapy buys months rather than days.

The most interesting scientific claim is that antagonistic receptor pairs are predictable from structural models of the tail fibre and the host receptor, so the arena confirms a design rather than searching blindly. That would make phage therapy a rational design discipline in the way antibiotics never became.

04Why now

AI-designed phage and FDA-cleared libraries make the supply problem tractable, and cheap sequencing makes the evolutionary readout routine. The field can now afford to select for evolution rather than for lysis.

05The decisive test

06What would break it

• Arena conditions may not reflect the infection site.
• Immune clearance may matter more than bacterial evolution in vivo.
• Payload-bearing phage raise their own regulatory questions.

07If it works

Phage therapy becomes a designed evolutionary intervention with a quantitative selection criterion, and resistance stops being an afterthought.