June 2026 thesis

Generation is cheap. Response is the frontier.

The interesting thing is not that AI can make content. It is that content can be generated, scored against a neural proxy, personalized, distributed, and revised as a system.

88%

organizational AI adoption

Stanford AI Index 2026

4 in 5

university students using generative AI

Stanford AI Index 2026

$5.4B

Suno post-money valuation after June 2026 Series D

Suno

20k

approx cortical vertices predicted by TRIBE v2 average-subject output

Meta / FAIR

The creative stack

AI media is becoming a closed loop.

01

Taste

The human chooses the frame: what should exist, who it is for, what emotion it should carry, and what should be killed.

02

Generation

Music, video, voice, images, code, analysis. Output cost keeps falling while fidelity and control keep rising.

03

Simulation

Models like TRIBE v2 make it plausible to score media against predicted brain response before a human audience ever sees it.

04

Distribution

Feeds, search, group chats, agents, creators, ads, and prediction markets decide which generated artifacts become reality.

05

Feedback

Clicks, trades, retention, revenue, memory, and behavior feed back into the next generation loop.

Why TRIBE v2 matters

The brain-response layer is no longer purely science fiction.

Meta's TRIBE v2 is a multimodal model that predicts fMRI brain responses to naturalistic video, audio, and text. The repo exposes inference against an average subject and projects responses onto a cortical mesh.

That does not mean you can perfectly predict culture. It does mean the loop is visible: generate a song, generate a video, simulate a coarse neural response, revise the stimulus, then push it into a feed where the real market tells you what lived.

Input

video + audio + text

Model

unified multimodal transformer

Output

predicted fMRI response

Surface

average-subject cortical mesh

Use

in-silico media testing

Risk

attention optimization without taste

Video frontier

A moving leaderboard, not a settled category.

Current with-audio text-to-video arena data from Artificial Analysis. Elo comes from blind user comparisons, not company demos.

ModelCreatorEloAPI price
1. Dreamina Seedance 2.0ByteDance1215$9.07/min
2. HappyHorse-1.0Alibaba-ATH1122$13.20/min
3. SkyReels V4Skywork1108$21.00/min
4. Kling 3.0 1080pKlingAI1103$20.16/min
5. Kling 3.0 OmniKlingAI1098$16.80/min
6. Veo 3.1Google1092$24.00/min
7. Sora 2OpenAI1088$6.00/min
8. LTX-2.3 FastLightricks974$2.40/min

Bits versus atoms

The old map still matters. It is just not the whole map anymore.

Bits are where the model wave lands first because the workflow is already digital. Atoms are where the money and friction live. AI media is a third thing: bits that simulate atoms, emotion, and social response.

Bits

Consumer creative tools

hot / crowded

The first wave was image generation. The current wave is video, music, voice, and personalized model surfaces.

Professional services

high spend / high competition

Legal, consulting, accounting, research, diligence, and analyst workflows are obvious because the inputs and outputs are already documents.

Finance and markets

incumbent-heavy

Fraud, underwriting, trading, prediction markets, research, compliance, and customer ops all become model-mediated.

Developer tooling

already colonized

Still one of the cleanest cases: technical users, immediate feedback loops, measurable productivity, and willingness to pay.

Atoms

Construction

large / under-digitized

Document workflows, bidding, RFIs, compliance, scheduling, and takeoffs are software problems hiding inside physical-industry culture.

Manufacturing

distribution gap

The top firms have AI programs. The long tail needs packaged, boring, ROI-positive systems that fit existing operators.

Government

slow / massive

Permitting, benefits, procurement, intake, 311, document review, and citizen services are enormous but procurement is the constraint.

Healthcare operations

messy / durable

Clinical AI gets attention. Admin, revenue cycle, coordination, staffing, and elder-care information layers may be more deployable.

Market map

Economic mass versus AI penetration.

Directional, not gospel. The point is to see where model adoption is obvious, where distribution is hard, and where the opportunity is large enough to survive messy implementation.

All Sectors — tap to expand

Working conclusion

AI makes the artifact easier. It makes the responsibility for taste, intent, governance, and distribution harder.

Meta / FAIR TRIBE v2

Open-source multimodal brain-response model for video, audio, and text.

Suno v5.5

Voices, custom models, and My Taste as personalization primitives for AI music.

Suno Series D

$400M+ raised at a $5.4B post-money valuation in June 2026.

Google DeepMind Veo

Video generation with native audio, realism, prompt adherence, and creative controls.

Artificial Analysis Video Arena

Blind-vote text-to-video rankings with audio, Elo, samples, and pricing.

Stanford AI Index 2026

Adoption, capability, data-center, and model-performance benchmarks.

a16z Top 100 Gen AI Apps

Creative tools shifting from standalone image generation toward video, music, and voice.