Yogesh Sharma

AI Engineer · Author · Founder

I build AI agents that do real work — and I write about where they're taking us.

I'm a forward-deployed AI engineer in Seattle. I sit with the people who have the problem, build the agent that solves it, prove it works, and hand it over. I'm also building two products of my own, Befundo and Fitted, and writing The Agentic Future, a book series on what agents mean for the systems, companies and society around them.

What I'm working on The Agentic Mind
Portrait of Yogesh Sharma

Forward-deployed AI

From a customer's problem to an agent in production

The interesting part of AI isn't the demo. It's the last mile: understanding a real workflow, deciding what an agent may and may not do, building it, measuring it, and leaving behind something a team can actually run. That's the work I do now.

Featured engagement

PriorAuth AI — a prior-authorization copilot

Prior authorization is slow, manual and expensive for clinics and payers alike. Modeled on a regional hospital network, the goal was an agent that drafts requests a reviewer can trust — and knows exactly when to hand a case to a person.

  1. DiscoverSat with clinical and revenue-cycle stakeholders to map the workflow and where time and denials were lost.
  2. ScopeWrote the Statement of Work: outcomes, success metrics, data access, and the cases the agent must refuse.
  3. ArchitectAgents read the chart note, retrieve payer policy and medical-necessity criteria, and draft the request with citations.
  4. BuildOpenAI Agents SDK, AWS Bedrock AgentCore and FastMCP tools, with role-based access, PII protection and audit logging.
  5. ProveEvery release gated on evaluations — LLM-as-judge, golden sets and regression tests — with observability on every run.
  6. Hand overA runbook and an escalation path, so the customer runs it without me.

A person decides

Anything that reaches a real person waits for one. Agents draft, propose and prepare; people approve.

Grounded, not invented

Answers cite their sources, links are verified before anyone sees them, and generation is held to facts it was given.

Measured before shipped

Evaluations decide whether a release goes out — not a good-looking demo.

Working with LangGraph, the OpenAI Agents SDK, CrewAI, AutoGen, MCP and A2A, RAG and vector search, AWS Bedrock AgentCore, and fine-tuning open models.

Projects

Two products I'm building

Befundo

befundo.ai ↗

One place for families and the people who teach their kids. Parents find classes, camps and tutors; the businesses that run them get scheduling, enrollment, payments, messaging and live online classes in one product — on the web, iPhone, iPad and Android.

A team of agents

Ask Befundo routes each question to one of six specialist agents — discovery, family, help, listings, teaching and more — each limited to the tools it is allowed to use.

A curriculum that checks itself

Over 2,200 skills in a prerequisite graph, with lessons and practice written by models and every answer key re-checked by an independent pass before a child sees it.

Growth with a human hand

Grow finds and scores prospects and drafts outreach and ads — created paused, sent one approved email at a time.

Fitted

thefitted.online ↗

A copilot for the job search. Fitted gathers postings from job boards and alerts, tailors a resume to each one from a single master profile, tracks every application, and fills in the forms — stopping before Submit, so you always decide.

Tailored, never invented

The model can only reference facts from your profile by id, so a tailored resume cannot claim experience you don't have.

You approve everything

Review and edit each tailored resume before it's used, and export it as PDF or Word.

The whole search in one place

Postings, applications and follow-ups tracked together, with a shareable profile and booking page.

Cover of The Agentic Mind, The Agentic Future Book One, by Yogesh Sharma

Just published · The Agentic Future, Book One

The Agentic Mind

From Agents to Agentic Infrastructure

Agents are moving from demos into the systems we run every day. The Agentic Mind explains how they work — memory and context, reasoning and planning, tools and actions, and the networks agents form with one another — and what infrastructure it takes to run them responsibly. It is written for builders and leaders alike.

Visit the book's website

On Amazon: Kindle · Paperback · Hardcover

Cover of The Agentic Enterprise, The Agentic Future Book Two

Coming next · Book Two

The Agentic Enterprise

From infrastructure to organizations, economics and society.

About

Twenty years underneath, now building on top

Before AI, I spent two decades on the infrastructure people never see — mobile networks and messaging at AT&T, T-Mobile's first container platform, F5's first cloud services, and Amazon Keyspaces, AWS's serverless Cassandra-compatible database, where my teams took availability to five nines. Most recently I led AI platform and agentic systems work at Optick Labs before going independent.

Running mission-critical systems taught me what agents will need to earn the same trust: clear limits, honest measurement, and a person in the loop. That's what I build, and what I write about. I'm a graduate of Stanford's LEAD program, once founded a STEAM program for kids in Redmond, and live in Seattle.

Let's talk

About agents, a problem you'd like one to solve, the book, or something you're building.

Connect on LinkedIn The Agentic Future