Director, Product Management , Agentic Optimization

Salesforce•San Francisco, CA

12d•$197,300 - $344,700•Hybrid

About The Position

Salesforce is the #1 AI CRM, where humans with agents drive customer success together. Agentforce is Salesforce's platform for building, evaluating, and managing AI-powered agents that transform how customers engage with brands and how work gets done. Agentforce Observability is the product surface where enterprises measure, monitor, and improve those agents in production — closing the loop between what agents do and how well they do it. We're hiring a Senior PM / Director to own agentic analytics for Agentforce — using AI to analyze how enterprises' agents actually behave in production, cluster failures at scale, and give customers concrete recommendations to make their agents better. This is one of the highest-leverage bets in the Agentforce portfolio: AI applied to the hardest problem in enterprise AI, which is knowing whether your AI is any good and what to do about it. You'll own the roadmap, the user experience, the go-to-market, and the design partner and pilot programs. You'll set the strategy that takes this product from a pilot with a handful of Salesforce's most sophisticated customers to a differentiated GA offering embedded in every Agentforce deployment. This is an IC role for a builder who thinks in agents. You use coding agents (Claude Code, Cursor, Codex, etc.) daily and have a point of view on where they're going. You prototype before you spec. You've built and evaluated agents yourself, and you have opinions about what makes agent optimization actually work — evaluation harnesses, annotation flows, failure clustering, and the feedback loops that turn raw session data into product intelligence.

Requirements

Product management experience, with a track record of shipping products for technical users (developers, admins, or technical SMEs). Director-level candidates should show a clear pattern of scope and impact beyond that.
Coding-agent fluency. You are a daily user of coding agents. You've built and shipped agents yourself. You can talk substantively about LLM orchestration, tool use, evals, and what makes an agent actually work in production.
Applied depth in evaluation and agent quality. You've worked on evaluation frameworks, annotation flows, failure clustering, or the equivalent — and you can reason about how to measure whether an agent is actually getting better.
Comfortable with ambiguity, biased for action. You turn fuzzy problems into sharp bets with measurable outcomes. You'd rather ship a rough thing and iterate than debate a perfect one.
Strong written and verbal communication. You can turn ambiguity into direction, rally a team around it, and hold your own with customers and execs.
Cross-functional operator. You work well across engineering, design, and adjacent product teams. You know when to lead, when to follow, and when to escalate.
Customer empathy at scale. You lead with the customer problem — not the feature request. You can synthesize signal across a design partner cohort, distill it into crisp product requirements, and communicate tradeoffs clearly to engineers, execs, and the field.

Nice To Haves

Engineering background, or the equivalent hands-on depth — comfortable reading code, reviewing APIs, prototyping in a notebook, and partnering with engineers on real trade-offs.
Direct experience with the current agent-building and eval landscape (Cursor, Claude Code, Replit, LangSmith, Braintrust, Langfuse, agent frameworks) and strong opinions on what they get right and wrong.
Experience with customer-facing pilot or design-partner programs for early-stage AI products.
Design sense: can turn messy systems into clean primitives with opinionated defaults.
Experience operating with autonomy in a highly cross-functional environment, including executive, technical, and go-to-market stakeholders.

Responsibilities

Own the product end to end
Set and own the roadmap, outcomes, and product strategy for agentic analytics — from current pilot through GA and beyond.
Own the user experience. Partner with design to build an agent that customers trust and reach for, not one they tolerate.
Own the go-to-market. Work with Enablement, Docs, Marketing, and Sales on positioning, launch, and adoption. Run tight feedback loops with design partners and early customers.
Set and defend success metrics (pilot activation, failure-cluster coverage, recommendation acceptance, agent quality lift) and use them to make real trade-offs. Kill what isn't working.
Prototype with coding agents to pressure-test ideas before writing a spec. A working demo beats a 10-page doc.
Dogfood the product constantly. Run it against real customer data. Break it. Eval it. Bring receipts.
Stay close to the code and the runtime. Read the traces. Understand the eval harness. Sit next to engineering during discovery.
Run the design partner program directly. Lead pilot calls, extract requirements, turn customer pain into shipped product.
Bring the customer voice into every product decision. Distinguish real needs from workarounds, one-offs, and badly framed requests.
Push a fast-paced environment. Ship the smallest thing that works, learn, iterate.
Coordinate with peer PMs across Agentforce — the runtime, the authoring surfaces, evaluation, and the broader Observability portfolio — so agentic analytics fits into the ADLC as one coherent story.
Communicate clearly with engineers, designers, execs, and the field. Bring a point of view and change your mind when the data says to.