Person
1 episodes · 24 quoted claims
AutoGPT was, as Maxim Bar Kogan describes it, the first really autonomous agent built on large language models, letting the model decide actions and run them in a loop through tools.
“They created the 1st, uh, as far as I know, 1st really autonomous agent running on LLMs, right? So agent that, you know, would let LLM dis not generate text, but decide what to do”
Bar Kogan says AutoGPT did not work well because models such as GPT-4 were not yet good enough.
“Now it granted, it didn't work that well, it was too early, the models were not good enough, uh, GPT four was not good enough.”
Bar Kogan says Claude Code became the first widely used autonomous agent, followed by Cowork and OpenClaw.
“We had, uh, cloud code, which became like the really 1st, uh, widely used autonomous agent, and then we had, like, co work and open claw.”
Bar Kogan says a large share of Anthropic's revenue comes from enterprises paying for Claude Code to do work developers used to do.
“Anthropic's revenue is coming from enterprises that are paying for cloud code to do, uh, a lot of the work that developers used to do.”
Onyx Security trains models and builds agents that oversee other AI agents.
“Number one is we train models and build agents that can oversee other agents.”
Onyx Security sells its agent-oversight capability as a product called the secure AI control plane.
“we basically productize it in a product that we call the AI control plane or the secure AI control plane.”
Bar Kogan estimates that over 50 percent of agent use in a typical enterprise is autonomous coding agents and assistants, and about 45 percent is low-code automations.
“over 50 is the autonomous, uh, coding agents and assistants in the average enterprise. Then probably 45, uh, is, uh, is those, uh, uh, low code automations.”
Bar Kogan says Claude Code grew rapidly among Onyx's customers and Claude Cowork is now growing even faster.
“we would see cloud code growing like fire in our customer base, and now we're seeing a cloud cowork growing even faster.”
Bar Kogan says limiting a software system's permissions is traditionally the first and most important security control, but autonomous agents are given a user's broad permissions.
“Because, and then, no matter what, even if it goes wrong, even if it's compromised, it can't, um, typically do any stuff that it was originally allowed to do.”
Bar Kogan argues that existing endpoint and API security tools lack the context to understand what flexible AI agents are doing.
“So a lot of these existing tools, they don't have the context to understand what these very flexible, unpredictable systems are doing.”
Bar Kogan says running a frontier agent to supervise every agent would cost customers more than the AI they are overseeing and would add too much delay.
“if I need to run an agent for every agent you're running as your security vendor, uh, you're gonna be paying for me more than you're paying for your AI, right?”
Onyx trains small models whose single job is to decide when a smarter agent should review an action, keeping cost and latency low.
“They're very small, they're, they almost can't do anything else other than be able to say, should I have a smarter agent? Look at this.”
Bar Kogan says most of Onyx's research and engineering staff come from an Israeli intelligence unit focused on math and cyber.
“Most of the people in our company, most of our research engineering, come from a unit in the Israeli intelligence, where we actually deal with math and cyber and the intersection thereof.”
Bar Kogan argues that if AI companies become ten-trillion-dollar firms, customers will want a non-vendor company overseeing AI, which he calls a market worth over 100 billion dollars.
“If you have AI companies that are 10 trillion dollar companies, we think you want a company that is not the vendor of the AI itself to oversee and help you control what the AI is doing.”
Bar Kogan describes Onyx Security as a two-year-old company that nonetheless attracted inbound interest from Fortune 10 and 20 customers.
“We're a two year old company, we're like a few people from, uh, you know, Pudan Math and Cyber.”
Bar Kogan says automated vulnerability research seemed twenty to fifty years away ten years ago, and that Mythos-level models arrived suddenly.
“if you took me ten years ago, automated vulnerability research looked like a a dream that would take 2050 years to to happen.”
Bar Kogan says that if another developer, such as a Chinese lab, releases a Mythos-level model, a restricted rollout would look like a mistake because companies would be left vulnerable.
“The problem is that if anyone gets to a mythos level model earlier, then in retrospect, it will look like a huge mistake.”
Bar Kogan says few large enterprises still ban AI agents, though some financial firms restrict which tools are allowed, and he thinks allowing many tools is the better strategy.
“I personally think that the companies that are gonna do well are the companies that are gonna allow a lot of different tools.”
Bar Kogan says the two primitives of 2026 AI, large language model foundation models and LLM-driven tool calling agents, have not changed much in recent years, helping Onyx keep its core technology focused.
“we're still building agents in pretty much the same way, where we have an LLM decide what are the tool calls that we're gonna make and generate those.”
Bar Kogan argues security buyers prefer an independent party over the vendor of a product to certify it is safe.
“you're not gonna trust the vendor of a product to tell you that this product is not gonna mess your environment.”
Bar Kogan says enterprises are unwilling to share historical agent behavior data with Anthropic or OpenAI because they fear it would be used for training, which gives Onyx a data advantage.
“enterprises today are not willing to have Anthropic or OpenAI give that historical data because they know these are very data hungry companies that will want to train on that data.”
Bar Kogan says models increasingly make errors from independent, semi-aware perspectives that may not align with the user's, and that this grows with model intelligence.
“Maybe just the the way it is that as you get smarter, you have more independent thoughts and and more, uh, you're more conscious”
Bar Kogan says Israel is becoming strong in AI, with companies building world models, AI infrastructure and chips.
“Most of the people in our company, most of our research engineering, come from a unit in the Israeli intelligence”
Bar Kogan says Onyx believes mechanistic interpretability, studying a model's internal weights and activations, will be at least part of controlling advanced AI.
“We believe that understanding the internal weights and activations, what is the internal structure, the mathematical structure of these systems, is gonna be at least part of the solution.”