No Priors

No Priors Ep. 43 | With Clara Shih, CEO of Salesforce AI

Elad Gil and Clara Shih on salesforce AI CEO Clara Shih on Data, Copilots, and Enterprise Adoption.

Elad GilhostClara ShihguestSarah Guohost
Dec 7, 202327m
Clara Shih’s path from Hearsay Social and Service Cloud to leading Salesforce AISalesforce’s open AI architecture: in‑house models, customer models, and third‑party providersEinstein Copilot, Copilot Studio, and the platform for prompts, actions, and agentsCurrent state and trajectory of generative AI adoption in large enterprisesCentral role of data readiness, data clouds, and zero‑ETL integrations for AITransforming customer service and sales workflows with generative AI (e.g., Gucci case study)Future of enterprise software: AI as the new UI, pricing, and startup opportunities

In this episode of No Priors, featuring Elad Gil and Clara Shih, No Priors Ep. 43 | With Clara Shih, CEO of Salesforce AI explores salesforce AI CEO Clara Shih on Data, Copilots, and Enterprise Adoption Clara Shih, CEO of Salesforce AI, explains how Salesforce evolved from early NLP and transformer models into a broad generative AI platform spanning every Salesforce cloud.

At a glance

WHAT IT’S REALLY ABOUT

Salesforce AI CEO Clara Shih on Data, Copilots, and Enterprise Adoption

  1. Clara Shih, CEO of Salesforce AI, explains how Salesforce evolved from early NLP and transformer models into a broad generative AI platform spanning every Salesforce cloud.
  2. Salesforce is pursuing an open architecture that combines in-house and third‑party models, wrapped in a unified Copilot and Copilot Studio platform for prompts, actions, and bring‑your‑own‑models.
  3. Enterprise adoption of generative AI is still early—"second or third inning"—with leaders already deploying production use cases while most are focused on getting their data organized and connected.
  4. Shih anticipates AI will fundamentally change enterprise UX and software development, shifting from hard‑coded flows to goal‑oriented agents, while business models must balance AI compute costs with clear, measurable ROI.

IDEAS WORTH REMEMBERING

7 ideas

Enterprise AI platforms must support a mix of internal, external, and customer‑owned models.

Salesforce uses its own models, enables customers to fine‑tune their own via Data Cloud, and integrates with Anthropic, Cohere, OpenAI, Google Vertex, and others so different customers can choose or delegate model selection based on their needs.

A unified copilot and agent platform is key to scaling AI across products.

Einstein Copilot and Copilot Studio (Prompt Builder, Action Builder, Einstein Studio) provide shared infrastructure for prompts, actions, and models, so every Salesforce cloud can quickly layer AI into workflows while reusing common services and trust controls.

Data readiness is the main bottleneck to broad enterprise AI adoption.

Most enterprises have fragmented data across multiple lakes and systems; Shih notes that Salesforce Data Cloud and zero‑ETL integrations with BigQuery, Databricks, and Snowflake are growing fast because companies must unify structured and unstructured data to power training and RAG.

Start with narrow, high‑value customer service use cases rather than boiling the ocean.

Many enterprises are successfully deploying generative AI first in support: unifying knowledge bases, using vector search and RAG, and giving agents reply suggestions grounded in articles and case history, which shortens handle time and improves customer experience.

AI will increasingly replace rigid UX flows with dynamic, goal‑driven experiences.

Shih describes prototypes like “Generative Canvas,” where the system dynamically assembles UI components and visualizations as users converse with an AI copilot, shifting product work from hard‑coding flows to specifying goals and constraints.

Business models for AI must clearly link higher compute costs to measurable ROI.

Because AI has real COGS, pricing needs to be simple enough for customers to understand without token calculators, while demonstrating outcomes such as lower support handle time and higher sales conversion to justify spend.

There is substantial white space for startups across models, tooling, and applications.

Shih sees opportunity in domain‑specific models (e.g., legal, medical), tooling for evaluation and observability, and novel applications that understand and integrate with existing data graphs, acknowledging that large platforms like Salesforce cannot cover every niche.

WORDS WORTH SAVING

5 quotes

AI is the new UI, or maybe Slack is the new UI for AI.

— Clara Shih

It’s early—probably the second or third inning—for enterprise adoption of generative AI.

— Clara Shih

Most companies, especially in the enterprise, as you know, their data is just all over the place, and so that’s kind of like step one.

— Clara Shih

The job of the software engineer and product manager and designer is gonna shift from prescribing the how to describing the why and the what and the goal.

— Clara Shih

Salesforce, like, we’re doing a lot, but we can’t do everything.

— Clara Shih

QUESTIONS ANSWERED IN THIS EPISODE

5 questions

How should an enterprise prioritize and scope its very first generative AI use case to balance impact with risk?

Clara Shih, CEO of Salesforce AI, explains how Salesforce evolved from early NLP and transformer models into a broad generative AI platform spanning every Salesforce cloud.

What governance and trust-layer mechanisms are most effective to keep enterprise data safe when using third‑party models?

Salesforce is pursuing an open architecture that combines in-house and third‑party models, wrapped in a unified Copilot and Copilot Studio platform for prompts, actions, and bring‑your‑own‑models.

How will the role descriptions and skill sets of product managers and software engineers change as agents take over more of the ‘how’ in software?

Enterprise adoption of generative AI is still early—"second or third inning"—with leaders already deploying production use cases while most are focused on getting their data organized and connected.

What are the trade‑offs between building domain‑specific internal models versus relying on general‑purpose external models for key workflows?

Shih anticipates AI will fundamentally change enterprise UX and software development, shifting from hard‑coded flows to goal‑oriented agents, while business models must balance AI compute costs with clear, measurable ROI.

For startups, where is the most defensible opportunity layer—foundational models, tooling, or domain‑specific applications—in a world of large platforms like Salesforce?

EVERY SPOKEN WORD

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome