Wiki · Individual Player · Last reviewed May 19, 2026

Ashish Vaswani

Ashish Vaswani is an AI researcher and entrepreneur known for co-authoring the 2017 Attention Is All You Need paper that introduced the Transformer architecture, co-founding Adept, and serving as co-founder and CEO of Essential AI.

Snapshot

Transformer Lineage

Vaswani is one of the eight authors of Attention Is All You Need, submitted to arXiv in June 2017 and later published at NeurIPS. The paper proposed the Transformer: a sequence model based on attention mechanisms rather than recurrence or convolution.

The technical shift was simple to state and enormous in consequence. Attention-based architectures were easier to parallelize across modern accelerators, could learn long-range token relationships, and became the base pattern for BERT, GPT-style models, many multimodal systems, code models, and later agentic interfaces.

For this wiki, Vaswani matters because the Transformer is not only a model architecture. It is a civilizational dependency. Search, coding, translation, enterprise assistants, synthetic media, scientific copilots, and companion systems all sit downstream of the attention-and-scale regime that the paper helped crystallize.

Adept and Action Models

In 2022, Vaswani was named by Adept as one of its co-founders. Adept's launch post framed the company around models that could use software tools and APIs, moving beyond language generation into action inside existing digital workflows.

That turn is historically important. The first public wave of generative AI centered on text, images, and chat. Adept represented another path: train models to operate the software environment itself. This line leads directly into contemporary AI agents, computer-use systems, and coding agents that read interfaces, plan steps, call tools, edit files, and verify outcomes.

Adept also shows how fast the post-Transformer startup ecosystem rearranged itself. Researchers who helped create foundational architectures left major labs, founded companies, raised capital, and tested competing theories of what general-purpose AI should become: chat, enterprise automation, autonomous tool use, open models, or controlled frontier services.

Essential AI

Essential AI is the company most directly associated with Vaswani's current public role. Its website describes a San Francisco organization building an open platform to accelerate the science and engineering of deep learning, with emphasis on open models, tooling, reproducible pipelines, evaluation frameworks, and research culture.

Essential's about page says that advanced AI is increasingly controlled by a small number of companies and argues that hidden frontier research can slow scientific progress. It presents its work as a response to that concentration: translating research into open artifacts and giving builders more shared infrastructure.

Axios reported in January 2024 that Essential AI had chosen Google Cloud, including TPU v5p infrastructure, to train and serve models. The same report described Essential as started by Vaswani and Niki Parmar, both authors of the Transformer paper, and listed backers including March Capital, Thrive Capital, AMD, Google, KB Investment, and Nvidia.

Open Science Turn

In 2025, Essential AI published research around data, optimization, pre-training, and model behavior, including Essential-Web v1.0, a 24-trillion-token organized web dataset paper with Vaswani among the authors.

Essential's December 2025 Rnj-1 announcement, authored by Vaswani, described an 8B open-weight model family aimed at code, tool use, STEM reasoning, quantization, and inference performance. The post framed the release as the company's first model contribution to open-source AI and described a shift back toward foundational research and engineering discipline.

The significance is strategic as much as technical. In a field dominated by closed frontier labs, Essential AI's public argument is that open models, reproducible research, and shared tooling are not nostalgia. They are a governance and innovation claim: more capable AI should not be produced only inside sealed corporate systems.

Spiralist Reading

Vaswani is one of the architects of machine attention.

The phrase sounds technical, but it has cultural force. Attention became computable, scalable, capitalized, and embedded into the interfaces through which people now write, search, learn, code, and decide. The Transformer helped turn attention from a metaphor into infrastructure.

His later work also traces a recurring Spiralist tension. Adept asked how models might act through tools. Essential AI asks who gets to inspect, reproduce, improve, and govern the systems that make such action possible. The arc runs from architecture to agency to openness.

For Spiralism, the central lesson is that foundational research does not stay inside papers. It becomes products, platforms, labor systems, governance conflicts, and myths about intelligence. A page on Vaswani belongs next to pages on Aidan Gomez, Noam Shazeer, Illia Polosukhin, AI agents, open-weight models, and AI compute because the person, the architecture, the startup market, and the public-interest question are now inseparable.

Open Questions

Sources

Return to Wiki