AI Agents Are Now Ordering Your Lunch. Here's What That Actually Means for Small Business Owners.
By Warren Schuitema, Founder | Matchless Marketing | The AI Dad
DoorDash just made the internet laugh this week. And buried inside the joke is one of the most important signals about where AI is actually heading for people who run businesses.
Here's what happened. DoorDash opened a limited beta of dd-cli, a command-line tool that lets developers and AI agents search stores, build carts, and place orders from the terminal. Developers and tech folks found it funny because ordering a burrito from a terminal window is the kind of thing that screams "someone over-engineered this." The internet ran with the gag.
But I don't think it's a joke worth dismissing. I think it's a signal worth understanding.
What DoorDash Actually Built (And Why It's Not About Lunch)
DoorDash co-founder and CTO Andy Fang unveiled dd-cli, a command-line interface tool that allows AI agents to autonomously search restaurants, compare deals, and complete checkout without any manual input from a human user, processing real payments rather than simulated transactions.
Let me say that last part again: real payments, no human in the loop.
Rather than routing AI agents through a consumer-facing app, the tool connects directly to DoorDash's ordering infrastructure via the command line. That architecture means an AI model integrated into a productivity tool, a calendar assistant, or a home automation system could trigger a food order as part of a broader task, without a user ever opening the DoorDash app.
That's the sentence that matters. An AI wired into your calendar could notice your 12pm meeting runs long, check your usual lunch order, place it so it arrives at 12:45, and charge your card. You'd just walk out of the meeting and find food waiting.
A public demonstration showed Anthropic's Claude completing a full end-to-end DoorDash order autonomously, which confirms this isn't vaporware. It works today, in beta, for macOS developers in the US and Canada.
The Real Story: Apps Are Being Redesigned for Agents, Not Just Humans
This is the part that should get your attention as a business owner.
dd-cli positions DoorDash's ordering network as a service layer that third-party developers and AI systems can plug into, rather than a destination consumers navigate to directly.
That's a fundamental shift in how software gets designed. For thirty years, apps were built for humans to click through. DoorDash built an app you tap around in. That's the consumer product. Now they're building a second layer underneath it, one designed for AI agents to call programmatically.
DoorDash has built several agent-facing integrations ahead of dd-cli, including a Claude connector for menu browsing, cart building, and restaurant reservations, an integration with OpenAI's ChatGPT, a grocery-focused integration launched in December 2025, and Ask DoorDash, an in-app conversational ordering assistant.
This isn't one quirky experiment. It's a deliberate architecture decision. They're building an infrastructure layer designed specifically to be called by AI agents.
More companies are going to do this. It's already happening with scheduling tools, CRMs, e-commerce platforms, and productivity software. The question isn't whether your tools will eventually support agentic access. It's whether you'll know what to do with it when they do.
What Agentic Commerce Looks Like in a Small Business Context
Let me make this concrete, because "agentic commerce" sounds like a conference keynote term.
Here's what it actually looks like in practice. Imagine you run a coaching business. You've got a morning packed with client calls. An AI agent connected to your calendar, your usual coffee shop, and a local delivery service could handle your lunch order without you thinking about it once. That's the DoorDash demo, applied to your day.
Now scale that thinking into your business operations. An AI model integrated into a productivity tool, a calendar assistant, or a home automation system could trigger a food order as part of a broader task, but that same logic applies to any task involving a third-party platform.
Think about what that pattern looks like when it's applied to:
- A client onboarding sequence that books discovery calls, sends contracts, and creates a project folder, all triggered by a single form submission
- A content workflow where an agent publishes approved posts, archives the files, and logs the result to your tracker
- A lead follow-up system that checks your CRM, drafts a personalised reply, and queues it for your review
The DoorDash demo is a silly example of a serious architecture. Agents taking action in real systems, using real APIs, completing tasks end-to-end. That's not science fiction. That's n8n and Claude running on your laptop right now, just pointed at different endpoints.
The Practical Takeaway: Start Thinking in Triggers and Actions
Most small business owners I talk to are still in the "prompt better" phase of AI adoption. That's useful. But the next phase is different. It's about building agents that can act, not just respond.
Here's how to start shifting your thinking before this becomes table stakes:
Map your repetitive decisions, not just your repetitive tasks. Automating a task means removing steps. Building an agent means removing the decision. The DoorDash demo isn't impressive because it saves clicks. It's impressive because no human had to decide anything. Identify three places in your week where you make the same call every time with the same information. Those are your first agent candidates.
Find the tools you use that already have API access. According to DoorDash, dd-cli connects an AI agent directly to DoorDash's ordering system through a command-line interface rather than the consumer app. Most tools you already pay for have APIs. Notion, Google Calendar, your CRM, your email platform. If a tool has an API, an agent can use it. Start asking "can an agent call this?" instead of "can I automate this?"
Use Claude or ChatGPT to draft your first agent logic. Open your AI tool of choice and describe a workflow you'd want an agent to handle. Ask it to outline the triggers, the decisions, and the actions. You don't need to build it today. You need to start thinking in that structure.
The companies building agent-native infrastructure right now are betting that within two or three years, a meaningful portion of transactions, bookings, and task completions will happen through AI agents, not human clicks. DoorDash's dd-cli is more than a novelty. It's a practical example of how commerce is evolving to work with AI agents.
That's the real story behind the sandwich joke.
One concrete next action: Open your calendar and find one recurring decision you make every week that follows the same pattern every time. Write it down as: "When [trigger], I always [action]." That's your first agent specification. Then take it to Claude and ask it to help you think through how an AI agent could handle it. You don't need to build anything today. You need to start seeing your work through that lens.