Newsletter

AI agents placed a few hundred of nearly a billion DoorDash orders

Agents barely order anything on DoorDash yet, Zwolle lets an agent pick from 12,000 purchasing categories, and a third had an AI policy in a poll.

By Piet Baudoin · 7 November 2026

AI agents are not taking over the easy work, but the work with many choices and many rules. Today: ordering through an agent, an agent in the purchasing team of a Dutch city, and the question of whether your office has rules for AI.

Are people letting their AI agent order yet?

Barely. DoorDash, a big American food delivery app, handled nearly a billion orders in the third quarter. CEO Tony Xu told investors last week: “AI agents are generating several hundred orders”. He called it a “very low use case”.

To be fair, the connection for agents has only existed for a month, for a small group of “early beta partners”. Xu does see agents in “long consideration purchases”: the purchases you think about for a long time.

What it means for you: you order a sandwich yourself; an agent helps first with the purchase that comes with quotes and terms. Read also what the agentic economy is.

Source: Ars Technica, 30 October 2026

Can an AI agent choose from 12,000 purchasing categories?

The first agent of the Dutch city of Zwolle is called Joop Inkoop (“inkoop” means purchasing). “Met meer dan 12.000 inkoopcategorieën helpt Joop Inkoop medewerkers bij het kiezen van de juiste cpv‑code”: with more than 12,000 purchasing categories, it helps staff pick the right CPV code, the European classification for public contracts. The software then creates the purchase request by itself. According to the city, the process is “volledig geaudit”, fully audited; the city council still has to set the rules for agents.

The advice from Zwolle and Wortell: “Durf te beginnen en keuzes te maken, ook als nog niet alles vooraf vaststaat.” Dare to start and make choices, even if not everything is settled up front. Wortell is a Microsoft partner and ran the training.

What it means for you: choosing from a long list by fixed rules is agent work, as long as someone can check the result. Read how to test an AI agent before you automate.

Source: Computable, 6 November 2026

Does our office actually have rules for AI?

At Wolters Kluwer Connections, a Dutch conference for accountants, “ongeveer een derde van de respondenten”, about a third of respondents, said their workplace has an AI policy. It was a poll of fifty to sixty attendees, so “De uitkomsten zijn niet representatief”: the results are not representative. Wolters Kluwer sells software itself.

AI expert Sanne Kanis offered three rules: there must always be a person who checks and signs off on the result, clients must know when AI is used, and organizations must record “welke AI-toepassingen en gegevens medewerkers mogen gebruiken”: which AI tools and which data staff may use.

That last one touches every office. Anyone using Gemini with a personal Google account reads from Google: “Please don’t enter confidential information that you wouldn’t want a reviewer to see”. A business Google account may come with different terms.

What it means for you: without a rule, every employee chooses for themselves, with their own account. Read why ChatGPT or Copilot alone is not enough.

Source: Accountancy Vanmorgen, 5 November 2026 · Google, Gemini privacy

Voices

An employee of a maker of point-of-sale systems, on ordering through an agent. Confirmed twice: others in the same thread say the same.

I work for a POS company and was asked about the ACP (Agentic Commerce Protocol) and I told my boss we should skip it for now because of this exact reason. Not anti-AI at all, in fact I've built tools for our clients to manage their menus via LLM

joshstrange · 5 November 2026

Someone who has AI read thousands of contracts, on the question that also belongs with Joop Inkoop. Confirmed once: the writer above has AI translate tender documents and also sees cases where it goes wrong.

Thousands of contracts, need to extract billing terms/logic. [...] The problem is, once I run it, can I trust the output without manual verification? How will I know if it skipped/missed something from a contract?

conductr · 5 November 2026

A software entrepreneur who found out what his paid personal subscription does with his chats. Confirmed twice: Google’s help page and several replies.

I learned this too late, thinking that having a paid subscription = not being turned into the product. Apparently paying is not enough.

jameslk · 4 November 2026

Where are we heading?

DoorDash handled nearly a billion orders in one quarter, and agents placed a few hundred of them (Ars Technica, 30 October). So the agent does not win at the pizza, but at the purchase you spend weeks on. I expect that within two years the first round of a business purchase at a growing company will run through an AI agent: a new supplier, an installation, a software package. Your agent puts the quotes side by side and reads terms, delivery times and the experiences of other customers. An agent does not see a banner and is not distracted by an offer that does not matter. It reads proof. In five years, I expect, selling works the same way: your company’s AI represents you to your customers’ agents, exactly as well as it knows you, in text and in images. The company that honestly shows what it is worth gets found on its own. And the order you enjoy placing yourself, you simply place yourself.

What does this mean for your business?

The common thread: an agent helps where many choices and fixed rules come together, and a person signs off. At Bombos that starts, for example, with orders that arrive by email. Bombos reads the email and the attachment and finds the customer. You approve. Our promise: the orders from your inbox are ready as jobs for your schedule, with the confirmation to the customer alongside. That way the same team does more work, and better work. That task is the start; after that your team teaches Bombos the next one itself.

Sources

Every source was opened on 7 November 2026 and every quote appears in it word for word.

Ars Technica (30 October 2026) · Computable (6 November 2026) · Accountancy Vanmorgen (5 November 2026) · Google, Gemini Apps Privacy Hub (undated) · Hacker News: joshstrange (5 November 2026), conductr (5 November 2026), jameslk (4 November 2026)

The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.