Hamming

The AI Conference 2026

San Francisco · September 29 to October 1, 2026

A private invite for The AI Conference attendees

Your agent passed the demo.Now let's see it survive real users.

Real users interrupt, go off script, and try to break things. Bring one production agent, voice or chat, to Hamming at Pier 48. We'll run adversarial and edge-case conversations against it live and show you exactly what fails.

Meet the Hamming team at Pier 48. Bring an agent your team owns.

Backed by Y Combinator. Built by AI leaders from

  • Tesla
  • Palantir
  • Meta
AI Conference agent teardown
San Francisco / Live

Examples of failures we surface

01

Tool calls

Retry loop triggered a duplicate transaction

critical
02

Prompt injection

Injected instruction overrode the system prompt

critical
03

Conversation

Interruption dropped collected context

high
Every failure saved as a repeatable test
  • 10K+ agents monitored

    across production systems

  • 50K+

    concurrent test calls

  • Trusted by

    Fortune 10 companies

  • Voice + chat

    conversational intelligence

How Hamming helps

Ship agents that hold up in production.

One loop from pre-release simulation and red-teaming to monitoring the conversations your agent has with real users.

01

Simulate before you ship

Run thousands of realistic conversations against your agent: interruptions, accents, off-script requests, and peak concurrency, before a release reaches users.

02

Red-team safety and security

Probe for jailbreaks, prompt injection, data leakage, policy bypasses, and unauthorized tool calls with adversarial agents that adapt mid-conversation.

03

Monitor every conversation

Catch regressions and emerging patterns across production calls and chats, then turn failures into the next test suite.

Your AI Conference agent teardown

Bring one real agent. Leave with a test plan.

Skip the slideware. We'll spend the time on your agent's behavior and the failures that matter for your next launch.

  1. 1

    Bring one agent your team owns: a phone number, a chat endpoint, or a recent production failure.

  2. 2

    We run adversarial and edge-case conversations against it and walk through what breaks.

  3. 3

    Leave with the highest-risk failures and a test suite you can run on every release.

Book a live agent teardown