# Aidan Gomez > 1995– · Computer Scientist, Transformer Co-author, Co-founder of Cohere > > **Recorded contribution:** Transformer co-author; co-founded Cohere — enterprise LLM company ## How to use this dossier Read for a causal chain, not a hero story: inherited problem → contribution → mechanism → downstream capability → limit. Then close the page and complete the reconstruction exercise from memory. ## 1. Historical orientation Canadian-British computer scientist Aidan Gomez was one of eight authors of the Transformer paper while interning at Google Brain. He later co-founded Cohere in 2019, focusing on language models and enterprise deployment. Gomez's contribution is shared Transformer co-authorship plus institution-building around production NLP; public accounts should not isolate an unsupported personal 'invention' within the paper. Gomez's Transformer co-authorship belongs to a collaboration that joined multi-head attention, positional encoding, residual connections, normalization, feed-forward layers, and a training recipe into one working translation system. His later Cohere work addressed a separate challenge: adapting language models to enterprise data, retrieval, security, and deployment constraints. ## 2. The problem inherited Sequence models needed parallel long-range interaction, and organizations later needed adaptable language models that could run within enterprise security, retrieval, and data constraints. ## 3. The central contribution Gomez co-created the Transformer and co-founded Cohere, helping move attention-based language models from research into enterprise platforms. ## 4. Reconstruct the mechanism 1. Represent tokens with content and position vectors. 2. Use multi-head attention to exchange information across the sequence. 3. Pretrain stacked blocks on large corpora with a language objective. 4. Adapt or ground the model through retrieval, fine-tuning, access controls, and task-specific evaluation. ## 5. What changed downstream - Transformers became the base architecture for contemporary language and multimodal systems. - Cohere contributed to competition and deployment patterns for enterprise language models. - The combined trajectory helped move Transformers from a research architecture into a general commercial platform for generation, embeddings, reranking, and retrieval-grounded applications. ## 6. Attribution, limits, and uncertainty - The Transformer is an eight-author result with extensive predecessors and later contributors. - Enterprise branding does not guarantee privacy, accuracy, security, jurisdictional compliance, or model fitness; company roles and capabilities are volatile. - Enterprise positioning does not establish confidentiality, accuracy, or return on investment; organizations must evaluate data handling, access control, citation support, latency, cost, and vendor dependence. ## 7. Reconstruction lab Implement one self-attention layer, then place it in a toy retrieval system over private documents. Add authorization filtering before retrieval and test for cross-user leakage. Build the same retrieval task with and without reranking, measure grounded answer support, and threat-model which enterprise documents an unauthorized prompt could expose. ## 8. Evidence trail - [Attention Is All You Need](https://arxiv.org/abs/1706.03762) — NeurIPS - [Aidan Gomez](https://cohere.com/about) — Cohere - [Aidan Gomez](https://en.wikipedia.org/wiki/Aidan_Gomez) — Wikipedia contributors --- *Research checked 2026-08-09. Dates, roles, and claims about living people are historical snapshots. Linked sources remain the authority; this dossier is original instructional synthesis.*