Towards Data Science

I recently covered TypesafeAI’s JEV in TDS (link at the end). JEV makes fast decisions, such as classifying content or choosing the next step in a workflow, and returns typed answers that software can use directly. The accompanying probabilities help an application decide whether to act on an answer or pass it to a person for review. The internet went a bit crazy over JEV, so it was no surprise to see rival products appear after its release.

While, conversational AI and agentic AI are now distinct foundational use cases of generative AI, moving an agent from a proof-of-concept to a reliable, production-ready enterprise workflow requires it to qualify on the measures of scalability, responsiveness, and cost-effectiveness. To understand how to achieve this, we first have to ask: Why are agents needed in the first place? Because conversational insight from data is often not enough. LLMs have the capability to plan, reason, and act.

AI adoption's main villain is safety. We've been seeing incidents of data exfiltration and confused deputy attacks. It happens because an LLM is probably the weakest link in the system. Most attacks happen when AI agents are exposed to the lethal trifecta. If agents can read from untrusted sources, access internal knowledge, and communicate to the outside world, they are vulnerable. LLMs can't differentiate between instructions and context. For the model, it's all part of the same prompt.

In September 2025, Horace He and colleagues at Thinking Machines Lab ran a simple experiment. They sent the prompt "Tell me about Richard Feynman" to Qwen3-235B 1,000 times at temperature 0 and asked for 1,000 tokens each time. Temperature 0 means the model always picks its most probable token, so you would expect 1,000 identical answers. They got 80 unique completions. What caught my eye is where the answers split. All 1,000 completions were identical for the first 102 tokens.

Your home for data science and AI. The world's leading publication for data science, data analytics, data engineering, machine learning, and artificial intelligence professionals.

I am very forgiving of an agent that is only talking. If it gets a draft or summary wrong, I crash out at my screen and ask again, and since nothing outside the chat window has changed, a retry is all it costs me. But the mood changes once the agent gets a tool. A wrong answer can now mean a sent email or a moved payment, and the usual fix, which is putting a second model in front to check the first, starts to feel like hiring an intern to supervise an intern.

AI applications The cost of control At Straight Up AI we’ve built an internal control plane for orchestrating coding agents. How much latitude agents are given is driven by three factors: Blast radius of a mistake. Mature production systems often have offline consequences whereas 0-1 MVPs do not. Project context. Legacy projects have less embedded knowledge that coding agents make probabilistic judgements on. Project maturity.

If you’re embedded in the world of software development, AI, and Large Language Models (LLMs), you probably talk about and use AI agents all the time. If you want to go beyond using agents and start to create your own, this article is for you. But before we build our first one, have you ever stopped to think about what an AI agent is? Let’s pin down exactly what we mean by the term — agent.

Agentic AI AI agents can now write more code than any of us can read. That changed how I think about AI. Image by author (used AI) AI agents now write more code than we can review. Here's why the real bottleneck is human attention, and how to hire AI colleagues using an org chart. Most of us still “use” AI. Open a chat, type a prompt, get an answer, close the tab. I think that era is ending. Here’s why. The agents have become productive enough that getting work out of them isn’t the hard part anymore.

If you work in product or tech, you have probably sat through this meeting. Someone forwards a vendor demo or a viral post, and suddenly the roadmap needs an agent, an MCP app, or a harness.