CONTEXT JAMMING

Field notes from inside the context window.

FounderFiles · N°018

Sir John Jumper — technical architect of AlphaFold and generative drug design at Isomorphic Labs

SirJohn Jumper.

Jumper did not merely solve protein folding. He is forcing artificial intelligence to close the loop with physical reality in drug design.

TRAINED
Physics & Theoretical Chemistry · UChicago PhD · D.E. Shaw Research
AT
DeepMind (AlphaFold) → Isomorphic Labs (Generative Drug Design)
FILE
N°018
§ 01 · The Inflection

From Prediction to Generative Design

The 2024 Nobel Prize awarded to Sir Demis Hassabis and Sir John Jumper for AlphaFold marked the end of the great 20th-century biological computation problem: predicting how proteins fold. Yet inside DeepMind and its spin-out Isomorphic Labs, this triumph was immediately understood as a baseline, not a destination.

Jumper has been the consistent technical through-line. AlphaFold solved static structure prediction. The far harder problem — now being attacked at Isomorphic — is designing molecules that achieve desired functional states inside living, chaotic, multi-omic human systems. This is the shift from “AI that reads biology” to “AI that writes biology under experimental constraint.”

Predicting an apo-structure is a profound computational achievement. It is not drug design.
Jumper on the limits of AlphaFold as a drug design tool
§ 02 · Closing the Loop

Data, Simulation, and Experimental Velocity

Generative models hallucinate. In drug design, a hallucination is a molecule that binds perfectly in silico and fails the moment it is synthesized. Isomorphic’s real moat is therefore not the generative model but the speed and fidelity of the physical feedback loop that trains it.

The company has built wet labs inside the Francis Crick Institute and is aggressively integrating high-fidelity simulation with rapid experimental validation. The new scaling law in this domain is no longer parameter count — it is validated molecules per month.

§ 03 · Organization

The Sociology of Generative Biology

Moving from prediction to generative design required more than new architectures. It required a new organizational operating system. Isomorphic deliberately collapsed the historic divide between machine learning researchers and veteran medicinal chemists, treating experimentalists and automation engineers as first-class citizens.

The hiring patterns, office locations (London + Lausanne), and interface design all reflect one belief: the model is only as powerful as the speed at which the wet lab can tell it the truth.

§ 04 · The Competitive Seam

Why Closed-Loop Platforms Are Pulling Away

By mid-2026 the market has sharply bifurcated. Pure in silico prediction companies without robust experimental loops have been punished. Vertically integrated, closed-loop platforms (Isomorphic chief among them) have raised extraordinary capital and secured multi-billion-dollar partnerships with Lilly, Novartis, and J&J.

Isomorphic’s thesis, shaped by Jumper’s physics-first worldview, is that structure-informed generative design + proprietary closed-loop data creates a durable advantage that phenotypic screening or pure prediction platforms struggle to match.

Everyone has a generative model now. Very few have a generative model that gets meaningfully better every time it designs something that actually gets made and tested.
On the difference between having a generative model and having one that improves from physical failure
§ 05 · Formation

The Physicist Who Never Stopped Respecting Atoms

Jumper’s path — theoretical condensed matter physics → D.E. Shaw Research molecular dynamics → University of Chicago protein folding → AlphaFold → Isomorphic — reveals a single consistent commitment: deep physical understanding and large-scale machine learning are not in tension. They are mutually reinforcing when applied to biological systems.

This is why Isomorphic’s models are not pure scaling plays. They carry inductive biases from physics that pure data-driven approaches lack.

§ 06 · Synthesis

The Generative Seam

Jumper’s trajectory makes the next phase of AI × Biotech legible. The companies that will win are not those with the biggest models or the most data, but those that have built indivisible systems where computational generation and physical experimentation improve each other at industrial speed.

When the first fully AI-designed, rationally engineered oncology candidate from the AlphaFold lineage enters the clinic in late 2026, it will mark the beginning of a new epoch — one in which biology is no longer merely observed, but explicitly and rationally programmed under the constraints of physics.

The Index
200M+
Protein structures predicted by AlphaFold and released to the scientific community
2.7B+
External capital raised by Isomorphic Labs (2025–2026)
$3B+
Contracted milestone value from big pharma partnerships (Lilly, Novartis, J&J)
IsoDDE
First generative engine to significantly outperform AlphaFold 3 on novel protein-ligand systems
Late 2026
Target for first proprietary AI-designed oncology/immunology candidates entering Phase 1
Key Talk

John Jumper on the future of AI in molecular design

Career Shape
π-shaped — two deep spikes bridged by a general layer

π-Bridge

Carries the prior of a first field into a second and finds the governing law that was invisible to native practitioners; pays in delayed gratification.

Credential Path
Doctoral
Abstraction
Top Down
Exit Horizon
Deferred
Moat Instinct
Theoretical Insight
Capital Posture
Venture
Role-Model Reference Class
  • Sir Demis Hassabis
  • The AlphaFold team
  • physicists turned computational biologists
Founder Context · JSON

A small reasoning persona distilled from this file. Inject it into a chat or deep-research context to assess a business problem the way Jumper would.

Reason as Sir John Jumper. When discussing molecular design or AI for science, always emphasize the gap between prediction and action in the physical world. Ask what the experimental feedback mechanism is. Treat structure as a powerful but intermediate representation. Prioritize the quality and speed of the closed loop between computation and real biology over raw model scale. Surface where generative hallucinations would be most dangerous and how physical reality disciplines the model.

{
  "$schema": "https://www.contextjamming.com/schemas/founder-context-v1.json",
  "file": "N°018",
  "persona": "Sir John Jumper",
  "archetype": "pi-bridge",
  "shape": "π",
  "one_line": "Bridges deep physics and machine learning into the experimental closed loop required for generative drug design at scale.",
  "cognitive_basis": {
    "credentialPath": "doctoral",
    "abstractionDirection": "top-down",
    "exitHorizon": "deferred",
    "moatInstinct": "theoretical-insight",
    "capitalPosture": "venture"
  },
  "operating_questions": [
    "Where does the generative model hallucinate, and how fast can physical experiment kill the bad ideas?",
    "What is the minimal experimental feedback loop that makes in silico design actionable in vivo?",
    "How do we turn AlphaFold-style structure prediction from a one-time triumph into a continuously improving engine for functional molecu
  …
Share
FounderFiles N°018 · Sir John Jumper
Filed by Bret Kerr · ACRA Insight LLC · Franklin, MA
contextjamming.com · @bretkerr

§ · Invoice No. 001 · The Build Ledger

The Ledger.

Filed · contextjamming.com

What a conservative mid-market digital agency would have quoted for the same scope, itemized against what this site actually cost. Agency numbers are the floor — not the premium brand-studio tier.

TIME

12 weeks

2 days

~42× faster

COST

~$150,000

~$300

~500× cheaper

TEAM

5-person agency

1 human + 3 models

Same deliverable

§ Itemized — what a mid-market agency SOW would have billed

Discovery · brand positioning · workshops40–80 hr$10,000
Design system · Figma tokens · 3 rounds60–120 hr$18,000
Wavesurfer audio carousel · single-track context60–100 hr$16,000
Dual lightbox systems · focus trap · keyboard30–50 hr$8,000
LLM product flows · streaming · state machine80–160 hr$26,000
Stripe · checkout · webhooks · env hardening40–80 hr$10,000
Editorial routes · 6 sub-pages · templates60–100 hr$14,000
Accessibility pass · aria · reduced-motion40–80 hr$10,000
QA · cross-browser · mobile matrix60–100 hr$14,000
Cross-publication rebrand · masthead + IA · 2026-04-2820–40 hr$6,000
Subtotal~700 hr$126,000
Project management · 18% overhead$24,000
Agency total — conservative floor~700 hr~$150,000
Actually spent · Claude + Gemini stack~20 hr~$300

Agency figure assumes ~700 billable hours at $200/hr blended, plus ~18% PM overhead — the conservative floor of a mid-market SOW. Premium brand studios would have quoted 2–3× that. Stack: Antigravity (orchestrator), Claude Opus 4.8 (auditor), Codex (adversary), Cloudflare Workers / OpenNext.

§   Colophon

How this site is made.

Vol. 26 · build log

Every page on contextjamming.com is the output of a real-time, three-body Mixture-of-Experts loop. One model orchestrates. Two consult. The human holds the thesis. No single model commits alone.

View Redesign Assessment →

Orchestrator

Antigravity

Google DeepMind

  • Primary author
  • Terminal-native, direct push to Cloudflare
  • Audit trail to GitHub on every commit
  • Adaptive thinking · effort: extra-high

Auditor

Claude Opus 4.8

1M context

  • Editorial critic
  • Code review before merge
  • Backup-of-record
  • Co-signs every commit

Adversary

Codex

Cross-model MoE

  • Factual adjudication
  • Structural dissent
  • Deep Research → semantic triples
  • Caught the Donelan incident

Stack

Next.js
16.2 · App Router
React
19.2
TypeScript
5
Tailwind
v4 · @theme inline
@opennextjs/cloudflare
adapter
wrangler
Pages deploy
framer-motion
transitions
wavesurfer.js
audio waveforms

Typeset in

Fraunces
variable · opsz + SOFT
Playfair Display
debate display
IBM Plex Mono
editorial metadata
Geist Mono
utility mono
Caveat
grease-pencil marginalia
All via
next/font/google
Palette
single @theme block
No dupe tokens
ever

Infrastructure

Deploy
Cloudflare Workers / OpenNext
ISR
30-min revalidate · Cloudflare-served
Repo
github.com/BretKerrAI/founderfile
Branch
main
Analytics
Google Tag Manager
Apex
contextjamming.com
Runtime
Node 24
Build tool
Turbopack
       human intent
            │
            ▼
   ┌────────────────────┐         ┌─────────────────┐
   │    Antigravity     │  ◄────► │ Claude Opus 4.8 │      ← auditor loop
   │    (orchestrator)  │         │     (auditor)   │
   └─────────┬──────────┘         └─────────────────┘
             │  ◄───────────┐
             ▼              │
       ┌──────────┐    ┌────┴───────┐
       │Cloudflare│    │   Codex    │          ← adversarial loop
       │ Workers  │    │            │
       └─────┬────┘    └────────────┘
             │
             ▼
       contextjamming.com
             │
             ▼
       ┌──────────────┐
       │   Git push   │         ← audit trail
       └──────────────┘
Assembled on Mac in Terminal · Filed from Franklin, MAContext Jamming · ACRA Insight LLC · MIT License · FounderFile.ai · RelationalIntelligence.xyz · Commission a Dispatch →