Not subscribed? Sign up to get it in your inbox every week.

This is a deep-dive feature, the second one after the Ordinal piece three weeks ago. Same deal as last time: I pick one tool because it makes a bigger shift concrete, and the shift is the point.

I used to lose whole afternoons inside Customer.io.

Building messaging workflows at Copy.ai went like this: open the visual builder, drag a trigger node onto the canvas, wire it to a delay, branch on an if/else, write the email, duplicate the branch for the variant, connect the exit conditions, then send yourself test after test to make sure the logic held. The strategy (what to say, to which customer, when) was settled before I ever opened the tool. Those afternoons were pure translation: telling a machine about a decision I had already made, one drag at a time.

That labor is what this week's deep dive is about, because a category of tools is quietly deleting it. The one I want to walk through is Omnisend, an email and SMS platform for ecommerce that has gone further than most: it is built agent-first, and the difference shows up everywhere from campaign setup to how you leave a competitor.

Omnisend ships an MCP server. MCP (Model Context Protocol) is the standard that lets an AI agent plug into a tool and actually operate it, and it is the same protocol I keep writing about in agent contexts. Connect Omnisend's to Claude or ChatGPT and the agent sees your live account: campaigns, segments, automations, results. No CSV exports, no tab-switching, no copy-paste between the dashboard and the chat.

Then you just talk. Ask for a re-engagement campaign for customers who bought twice but went quiet for 90 days, and the agent builds the segment and drafts the campaign against your real data. Ask it to adjust an automation's timing, and it adjusts the automation. The nine ready-made prompts on their page run from "give me a snapshot of my store's performance" to a full account audit that flags the automations you should have and don't, like an abandoned-cart series or a post-purchase flow, with prioritized recommendations.

Sit with the contrast for a second. The work I did at Copy.ai was translating decisions into node graphs by hand. Here the translation layer is gone: you state the decision, and the agent does the dragging. The afternoon becomes a sentence.

The corollary I care about just as much: analytics stop being archaeology. The old motion for "why did email revenue dip last month" was five dashboard tabs, a CSV, and a pivot table. Against a live MCP connection it is a question. Deliverability health, week-over-week comparisons, which subject lines pull, which automations actually make money: all of it is now conversational, with the agent doing the joins. Reporting used to be a skill position on the marketing team. It is becoming a prompt.

The second thing Omnisend does with its agent-first architecture is aimed at its competitors, and it is the detail I would study if I ran a SaaS company.

Switching email platforms has always been miserable in a specific way: the value in your old account is dozens of automations full of conditional logic, timing rules, and templates accumulated over years. Rebuilding that by hand on a new platform is weeks of exactly the node-dragging I described above, which is why most teams grumble about their platform at renewal and then stay. The setup pain was the moat.

Omnisend attacks the moat directly: a full migration in five business days, free, on any plan. Contacts land in the first day; the team rebuilds every flow with its conditional logic and timing intact, plus segments, forms, and templates, then verifies your sender domain and tests the flows before handoff. Five days, at zero dollars, for what used to be a quarter-long professional-services project.

An offer like that only pencils out if the rebuild itself has gotten cheap, and that is what agent-first buys them: a platform an agent can operate is a platform an agent can populate, so translating a competitor's workflow export into native automations is exactly the kind of structured work agents are now good at. My bet is the five days will keep shrinking, and the floor will be set by the one step that resists automation: a human verifying the migration actually worked, flow by flow, before real revenue rides on it. The checking is the bottleneck. Every agent story lands there eventually.

The strategic read matters more than the feature. For twenty years, SaaS vendors could treat implementation pain as retention: leaving hurt too much, so nobody left. The first agent-first platform in a category gets to weaponize that, absorbing the switching cost as a customer-acquisition expense. Expect the five-day migration to become a standard play in every category with painful onboarding, which is most of them. If your favorite tool's lock-in is made of setup effort rather than ongoing value, its moat is evaporating right now.

Zoom out, because Omnisend is one example of something much bigger.

The first era of SaaS put software in the browser and left the operating to you. Every platform came with an interface you had to learn and a translation tax you had to pay: your intent, converted to clicks, one afternoon at a time. The entire implementation industry (onboarding teams, admin certifications, agencies that configure other people's software) is built on that tax.

Agent-first is the second era: the software still runs in the cloud, but agents do the operating. This is exactly why every frontier lab is racing toward coding agents. Code is the universal way to operate machines, so an agent that codes can operate anything with an API, and every MCP server published makes another platform part of that operable surface. The email platform happens to be where I felt the shift first; the same conversation is coming for your CRM, your support desk, your billing system, and your data stack, in whatever order their vendors ship MCP servers.

The implication for your team runs through the same logic: a lot of knowledge work gets dramatically better when the implementor can code, or can direct an agent that does. The marketer who can tell an agent precisely what to build ships in an afternoon what used to wait in a queue behind a technical colleague. Which means the scarce skills just moved up a level. Translation (knowing where the buttons are, holding the admin certification, being the person who knows how the workflow builder works) is depreciating fast. Judgment about what the workflow should do, and the precision to specify it, appreciates just as fast. Hire and train accordingly.

This also extends the questions I have been giving you all month. Three weeks ago the vendor question was who owns your outcome after the signature. Two weeks ago it was what walks out with you if you leave. The agent-first question completes the set: can my agents operate this? A tool with a real MCP server multiplies your team; a tool without one is betting you will keep paying the translation tax forever. Put "where is your MCP?" in your next vendor evaluation, weighted heavily.

Take the tool your team burns the most clicking-hours in. Search its name plus "MCP." If a server exists, connect it this week and start with the read-only work: ask the agent for an account audit, the gaps, the numbers nobody has pulled. You will learn the state of your account and the state of the vendor at the same time. If no server exists, ask your rep when one ships, and treat the answer as roadmap signal about whether that vendor understands the era it is entering.

Then look at the tool you have wanted to leave for two years and check whether an agent-first competitor will eat the migration for you. The exit got cheap while nobody was looking.

The software is finally ready to talk. Say something.