Keith Peiris and the trouble with 25 million users

Twenty-five million people had tried the thing Keith Peiris helped build. For a presentation company, it was an excellent opening slide. The awkward question came afterward: who needed it badly enough to keep coming back? Tome could turn a prompt into a deck. Peiris and his colleagues had wanted something more exacting: a tool that helped people doing serious work explain difficult ideas. A great many sign-ups had brought that ambition into view without resolving it.

The decision that followed gives his career its present shape. He and co-founder Henri Liriani moved away from Tome’s presentation product and rebuilt around customer relationships. Their current company, Lightfield, makes an AI-native CRM. That abbreviation promises the orderly management of customer relationships. Anyone who has ever forgotten to update a sales record knows how much hope can fit inside three letters.

For Peiris, the move carries a particular irony. He had spent years building ways for people to tell stories. His next task was to recognize when the story about his own product needed a different ending.

The family business came before the pitch deck

His first commercial education took place in Ontario. As a child, he experimented with Photoshop and websites, then began receiving requests for work through online communities. His father, who worked in IT sales, spotted a business where his son had found a way to finance toys and games. They set up an office in the basement. Keith handled the making; his father took on sales and marketing.

There is something usefully unromantic about that arrangement. A website had to please a customer. A price had to make sense. Somebody had to ask for the business. Long before Peiris worked on products with enormous audiences, he had encountered the smaller, less forgiving unit of commerce: a person expecting something in return.

He studied nanotechnology engineering at the University of Waterloo from 2006 to 2011. This was a substantial detour from arranging pixels. Peiris belonged to the core of the university’s student nanorobotics group, and in 2010 he wrote about its third-place finish at the NIST Mobile Microrobotics Challenge. The team competed against research universities with a robot named EMMA, short for ElectroMagnetic Microrobotic Actuation.

One event was a two-millimeter dash. Another required moving tiny pegs into holes. The robot had to fit within 600 micrometers in every dimension. An image-recognition system guided the magnetic field controlling it. The scale was comic; the engineering was perfectly serious. Peiris’s account described a self-directed undergraduate group holding its own against more established competitors.

Waterloo also gave him an arena for making an argument. In November 2007, he and a fellow nanotechnology student won the Sandford Fleming Foundation debate final. The winners received $300 apiece. Engineering, persuasion, a cash prize: several later interests were already sharing a room.

A whiteboard, a co-founder, a better page

Peiris went on to product work at Facebook and Instagram. At Instagram, his responsibilities included messaging and camera augmented reality; his professional biography credits his product work during Instagram Direct’s growth to 500 million monthly active users. Later came head-of-product roles at Glossier and Citizen. Consumer products had given him repeated opportunities to see what people actually did with tools intended for them.

The next company began with room to explore. Greylock investor Seth Rosenberg, who had worked with Peiris at Facebook, remembers buying a whiteboard with him at Staples. They spent months testing ideas and building prototypes. Rosenberg’s account emphasizes helping Peiris identify what interested him, and introducing him to Liriani. A future company was still a question someone could erase.

Tome, founded in 2020, came out of a familiar workplace irritation. Personal communication had acquired cameras, video, expressive overlays, and the ease of sharing from a phone. Professional ideas were still being pressed into slides. The founders wanted pages that could accommodate live information and interactive work without demanding a small design project every time somebody had a point to make.

“We’re all storytellers.”

Keith Peiris, 2022

That belief was broader than a complaint about PowerPoint. A product team needs to convey a customer’s frustration. A designer needs to show a working prototype. A founder needs to explain a business. Flattening all those things into screenshots makes the document easier to send while sometimes making the idea harder to understand. Tome tried to give those materials a more accommodating home.

By February 2023, it had passed one million users. Its reported time to that milestone was 134 days. A $43 million Series B, led by Lightspeed Venture Partners, followed. The funding announcement put cumulative capital raised at $81 million. AI could now help generate the presentation itself, including text, images, and layouts. The blank page had acquired a very energetic assistant.

The attractions were obvious. Start with a few words; get an artifact substantial enough to edit. The danger was subtler. A plausible presentation can resemble completed work before it has earned that distinction. The audience, the purpose, and the facts of the business still matter. Those requirements do not become optional because the software has found an attractive arrangement of boxes.

The missing material was the business itself

As Tome expanded, the team explored more specific professional uses. Work with sales and marketing customers revealed a difficulty upstream of writing: the useful information was scattered. An email carried one detail. A call carried another. A CRM field might contain an older version of both. Generating a persuasive page could require reconstructing the business relationship first.

That became a different company’s assignment. Peiris’s explanation of the pivot centers on context: knowing enough about the person presenting, the audience, and the relationship between them. A stronger model alone would not automatically supply those missing particulars. The team followed the problem into sales workflows and eventually into the system where customer knowledge was supposed to reside.

The retreat was substantial. Tome had employed about 70 people; the pivot left a small team rebuilding. In his later telling, Peiris described switching off the old product before the new direction was fully settled. Board support mattered because the decision consumed the option of comfortably treating the new work as an experiment alongside the old business.

Office space became one practical asset. Peiris described offering desks to young companies willing to use the unfinished CRM and provide feedback. It was a direct answer to an early company’s problem: getting close enough to customers to discover what needed fixing. A user could be down the hall. Product feedback had a pleasantly short commute.

A customer relationship, with a memory

The connection to his earlier work becomes clearer in Peiris’s writing. In February 2026, he argued that models understood complex customer relationships better when given chronological narratives alongside the underlying information. A relationship changes over time. Yesterday’s objection can become today’s opening after a different problem is explained.

His team had experimented with graphs and structured representations, then moved toward continuously updated stories about people and companies. Peiris presented this as a finding from building Lightfield, rather than a universal rule for every AI application. The distinction matters. Customer relationships contain explanations, changing priorities, and exceptions that can be difficult to preserve in a static field.

There is a pleasing continuity here. Tome had been concerned with how to put an idea on a page so another person could follow it. Lightfield asks how to assemble a business relationship so a person or an AI agent can act on it. In both cases, arrangement changes what the information means.

His January writing supplied a more prosaic example. A customer tried asking the agent to migrate CRM data. The attempt showed promise but also errors. The team responded by adding code execution and working on APIs that let the agent modify the destination’s structure. Peiris saw an opportunity to reduce the work of switching systems. He also described the tedious import process his own company had been helping customers through.

In May, Lightfield added agentic pipeline generation, drawing on previous deals and customer language to support outbound work. This widened the ambition from remembering business activity to helping perform it. The founding mission remained tied to understanding customers as a company grows and the conversations multiply beyond any one person’s memory.

The announcement, then the ordinary work

On September 9, 2026, Peiris announced a $47 million Series A led by Andreessen Horowitz. More than 5,000 companies had signed up since Lightfield’s November launch, he wrote. The money would support growth and expand the work its agent could perform. Signups describe interest; the company still has to turn that interest into dependable use.

Peiris was already using Lightfield to think through decisions inside Lightfield. Before a September interview, he explored staffing scenarios, including what might happen if the company hired different numbers of sales representatives. It is a revealing use for a CRM founder: asking what a business might become, with the company’s accumulated information available to help reason about it.

October brought less theatrical work. On October 16, he and Liriani published improvements to sequence templates, collaborative editing, and related-record creation. A week later came mail and meeting classification, fuller LinkedIn conversation syncing, and recorder changes. The release notes describe the small frictions that sit between a broad promise and somebody finishing a job.

That is where this chapter leaves him. Peiris has built websites for customers, helped tiny robots move, worked on messages exchanged by millions, and tried to make professional storytelling easier. Now he is working on software that can remember enough of a customer’s story to be useful tomorrow. After an audience of 25 million, tomorrow is a demanding place to begin.