Every pitch for agentic commerce opens with the model. Which frontier lab it runs on, how smart it is, how many PhDs sit behind it. This episode argues the model is the part that matters least.

ㅤ

Hosts Matt Johnson and Floyd Blaikie of Pivotree's Data vs. Commerce sit down with Anatolii Iakimets, Director of Product Marketing at Kibo. Anatolii takes the data side: an agent is a model plus a harness, the model is becoming a commodity, and the real work sits in the data and integration layer underneath.

ㅤ

The friction is clean. Floyd keeps reframing agentic commerce as a familiar platform decision, agents in a trench coat. Anatolii keeps pulling it back to the data. Commerce is deterministic. The price and the tax and the T-shirt size have to be exact, and an agent is only as good as the structured data it can reach and read.

ㅤ

👤 Guest Bio

Anatolii Iakimets is Director of Product Marketing at Kibo Commerce, where he focuses on B2C commerce. He has spent more than a decade in commerce and telecommunications, with earlier roles at Bold Commerce, Elastic Path, and Netcracker, and he has worked hands-on with AI and machine learning since around 2015. On this episode he takes the data side, arguing that agentic commerce succeeds or fails on data quality and integration, not on the model you pick.

ㅤ

📌 What We Cover

• The two parts of any agent: the LLM and the harness, the code that tells the model what to do
• Why the model is becoming a commodity, and why the switching cost for users stays low
• Why coding and text agents tolerate a slightly different answer every time, and commerce does not
• Why the price, the tax, and the T-shirt size have to be accurate to the point, not "good enough"
• Why an agent has to be integrated like an application, talking to your systems through MCP or an API
• How burning tokens turns into a real budget problem, with Uber's four-month AI burn as the warning
• The three things to weigh before you buy: composability, simplicity, and the ability to start small
• Why the "explain" function is the one customers reach for first, and the risk of a three or five year lock-in

ㅤ

🔗 Resources Mentioned

• Kibo (the guest's company)
• Anthropic (Claude) and OpenAI (ChatGPT) as frontier model providers
• Google as a frontier lab
• DeepSeek and Kimi K2 as open-weight models
• Model Context Protocol (MCP), APIs, and agent-to-agent protocols
• Microsoft Copilot's shift from subscription tiers toward usage-based pricing
• Uber's AI budget story
• Sam Altman on models becoming "intelligence on a tap"