Real Alpha
Watch modeSubscribed

MTS · 3 min · Oct 20, 2026

Agent platforms need built-in workflows and human approvals

Marco Palladino lists durable workflows, approval gates, distributed locking and database branches as built-in capabilities. They help agent products handle long-running tasks and avoid duplicate runs.

Marco Palladino

Marco Palladino is cofounder and CTO of Kong.

Co-founded Kong and launched Volcano, an AI-native platform for building and running production AI agents.

  • Theo Jaffee, Host · Co-founder & host, MTS

    One of the first members of a16z's new media team; earlier hosted the Emergent Ventures-backed Theo Jaffee Podcast.

  • Sophia Dew, Host · Host, MTS

About this segment

Marco Palladino lists durable workflows, approval gates, distributed locking and database branches as built-in capabilities. They help agent products handle long-running tasks and avoid duplicate runs.

From “AI Agents Will Automate Entire Enterprise Orgs | Marco Palladino”, Oct 20, 2026. Watch the full episode

“Being able to then have human approvals out of the box, being able to implement capabilities like distributed locking”
Marco Palladinoat 12:12 · MTS12:12

Takeaways

  1. Workflows may wait days, months or a year for new data.
  2. Distributed locks prevent duplicate runs for an execution.
  3. Database branching and user management are built in.

In this episode

Full episode
  1. 4:49AI success shows up in outcomes, not token savings3:28
  2. 8:56Agents need infrastructure beyond a typical SaaS stack2:58
  3. 19:49Workers will shift from doing tasks to orchestrating agents2:49

  4. 11:59Agent platforms need built-in workflows and human approvals2:31

Up next · MTSCustomer demand convinced Daytona to pivot to agents3:09

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

  1. 12:04

    So if you just like walk us through that, what does that mean? Well, so our agents are going to run durable workflows, workflows where we have to do some operations, maybe wait for like a day or even a month or even a year for new data to come in. Being able to then have human approvals out of the box, being able to implement capabilities like distributed locking so that if we have something that triggers the agent, it only runs once for each type of execution.

  2. 12:29

    So we don't have multiple execution of agents competing with each other, the ability to branch databases, the ability to also cater to an unlimited number of users that may be triggering the agent. So the agent is an application. Then on one end, we have the end user that may be triggering the agent. And then as it triggers the agent, the agent goes and does the durable workflow that it needs to do.

  3. 12:50

    Now, those users, they need to sign up. We need to keep track of them. We want to be able to manage that. So Volcano builds in unlimited user management. So whether we have one user or a million users, it's completely built in and unlimited. And then we can go and cater agentic products that are agents since day one. Remember a few years ago, when we started Kong, by the way, people were asking us, what is an API?

  4. 13:15

    And today it's clear that APIs are the backbone of the internet. And when you think about LLMs, they're also powered by APIs. So APIs became truly the number one backbone component that powers the whole world. And people were asking us, what is an API? And then slowly, we started to see API-first businesses. Think of Twilio. The product is an API. Think of Stripe. The product is an API. Think of AWS, the cloud.

  5. 13:41

    Effectively, the product is an API, an API to create a server, an API to provision a container. Now, as this new agentic opportunity finds its footing, we're going to be having agent-first type of products where the product is the agent. And we're seeing that already with the chats, the chat GPTs, and so on of the world. So as we go and build agent-first products, what we need is a platform that can cater to the very unique requirements that these agents are building, all the things we discussed earlier.

  6. 14:09

    And so we think that with Volcano, we're doing that and we're simplifying the creation of these agents, you know, because obviously agents are like products. The more we create, lots of iteration, something works, something doesn't. We don't want to worry about the platform. We want to worry about the distribution and the use case we're building.