The Web Form Won't Survive the Agent Era

A web form is a paper form wearing a digital costume. Someone digitized the printed intake sheet twenty-five years ago, added a submit button, and called it innovation. The form never got smarter. It just moved online, and it has stayed almost exactly the same since.

Every field on that form gets treated as intent. Fill in a job title and a half-dozen automated cadences start firing from tools that don't talk to each other. The visitor wanted an answer to one question. Instead they get a drip sequence built by a marketing operations team that has never met them.

Marketers argue about how many fields a form should have and whether the email field should require a business domain. I do the opposite of what most of that debate assumes. I sign up for newsletters with my personal email, on purpose, because I read it on weekends and my business inbox buries anything that looks like marketing. Half the posts I write about software companies started as a newsletter that landed in my own personal inbox, not my business one. The form didn't route that intent anywhere useful. My own habits did.

Every Field Is a Guess, and the Guess Costs You Visitors

Formstack's 2025 research, reported by marketing platform Zigment, found that form abandonment climbs past two-thirds once a form asks for more than seven fields. That is a vendor-reported figure, not an audited industry standard, but the direction matches every other study on the subject: more fields, fewer completions.

Cutting a form to three plain fields raises completion. It also hands the sales team a contact record with no budget signal and no way to tell a decision-maker from an intern doing competitive research. Shortening the form doesn't fix the underlying problem. The underlying problem is that a form can only collect what it was built to ask for, and it was built to ask for structured fields because nothing on the other end could handle anything else.

Drift Tried to Kill the Form a Decade Ago

Drift built an entire category around "get rid of forms" starting around 2016. The pitch was right and the execution was early. Decision-tree chatbots got tedious fast, visitors learned to click past the greeting bubble, and Drift itself became part of Salesloft's platform after an acquisition in 2024. The idea didn't die. It waited for a model that could actually hold a conversation instead of routing people through a script dressed up as chat.

The form didn't route that intent anywhere useful. My own habits did.

That model exists now. A visitor can type or say what they're trying to solve, and a language model can turn that into the same structured record a form was built to produce: name, need, timeline, budget range. The difference is the visitor gets a useful answer first and gives up their information as a byproduct of getting help, not as a toll paid before help arrives.

Two Numbers Beat Twenty Fields

Here's what I'd test before I'd test anything else: ask for one thing, a phone number or an email, and let the choice itself tell you something. A visitor who hands over a phone number is signaling higher intent than one who hands over an email, because a phone number invites a text message back, and a text message is much easier to answer than reopening a form on a desktop browser, granting microphone permission, and typing into a dozen fields.

Text-based qualification already exists as a product category. SMS platforms now run structured intake over text, gathering budget, authority, and timeline through a short back-and-forth instead of a form, then routing only the qualified contacts into a CRM. Voice memos work the same way on mobile: a visitor records thirty seconds of context, a model transcribes and structures it, and a rep opens the conversation already knowing what the person needs.

Malbek's Real Lesson Isn't About Contracts

Malbek sells contract lifecycle management, not marketing lead capture, so this isn't a pitch for their product. It's a pitch for their architecture. Malbek's extraction layer processes agreements paragraph by paragraph and builds a taxonomy running into thousands of data points, instead of the fifty to two hundred fields a standard system captures. Its assistant, Bek, takes typed prompts or voice commands and handles the retrieval and drafting itself, rather than making a user click through menus to find the right template.

Disclosure, since I'm writing about their architecture: I'm keynoting Malbek Envision in Austin this October. Take that as full disclosure, not distance from the point above.

That's the model marketing intake needs. Capture the unstructured version first, whatever someone actually says or types. Structure it after the fact. Route only the qualified, structured result into the CRM. The form does this backward: it forces structure before it has any idea whether the person filling it out is worth structuring for.

SoundHound Is Building the Pipes, Not the Landing Page

SoundHound's enterprise business grew fast through 2025 and into 2026, with its agentic platform handling conversational interactions across phones, kiosks, drive-thrus, and in-vehicle systems for customers in financial services, healthcare, and retail. The company introduced an agentic voice commerce platform at CES 2026, aimed at in-car and at-home shopping through voice assistants. None of that was built for marketing lead forms. What it proves is broader: enterprise buyers already trust voice and conversational channels to complete transactions that used to require a screen and a keyboard. A marketing team waiting for a dedicated "kill the form" vendor is waiting for a product that may never ship, because the infrastructure for voice-first and text-first capture already exists inside companies solving adjacent problems.

The Real Constraint Is the Agent Reading Your Screen

Agentic browsers already navigate sites and fill out forms on a person's behalf, and adoption is climbing fast. Form filling is also one of the tasks these agents handle worst, scoring lowest among write-heavy tasks on Skyvern's open Web Bench evaluation. A ten-field demo request form built for a human with a mouse is a bad interface for a human, and it's a worse interface for an agent trying to act on that human's behalf.

This is where a frontier lab's contribution isn't a marketing product at all. Anthropic introduced the Model Context Protocol in 2024 as an open standard for how an AI agent connects to external tools and data, then handed it to a neutral, Linux Foundation-backed body so it wouldn't stay any one company's property. The relevant idea for marketers: an agent acting for a buyer can hand over verified, structured intent through a protocol instead of guessing at which box on a form means "job title" and which means "company size."

No Single Vendor Kills the Form

Malbek didn't build a marketing product. SoundHound didn't build one either. Anthropic isn't going to ship a "delete your forms" button. The shift is happening across three separate layers that nobody has bundled yet: a capture channel that takes voice or text instead of typed fields, an intelligence layer that turns that unstructured input into the structured record a CRM expects, and a protocol layer that lets agents hand over verified intent instead of filling in blanks on your behalf.

A marketing team waiting for one company to ship all three will wait years. A team that wires a text-first intake to a model that scores intent, and points that pipeline at whatever CRM already runs sales, can retire its longest form this quarter.