a16zWhy the Next Generation of Enterprise Software Looks Nothing Like Salesforce
a16z
Introducing Lightfield and Its Founder 0:00
This episode features Keith Paris, CEO of Lightfield, joined by hosts Joe Schmidt and Alex Rampell. Lightfield recently raised a $47 million Series A led by the hosts' firm, and the company describes itself as building a business world model, a system that turns customer emails, calls, and meetings into a record that AI agents can use to get work done. One early example shared is a customer called Power, which runs a marketplace connecting people with illnesses to frontier medical treatments, and Lightfield reportedly helped someone with Alzheimer's find such treatment within days.
From Consumer AI Presentations to Explosive Growth 1:32
Keith explains that his team started as consumer-focused builders, having worked on selfie design at Instagram and Messenger, and believed large language models would change how people communicate. They built a presentation tool that used GPT to generate slides, launching it around the same time as ChatGPT and GPT3. The product grew explosively, reaching about 2 million users a month, with demand so high they didn't have enough inference capacity to support everyone.
Deciding to Abandon a Successful Product 2:30
Despite the growth, Keith says the team never actually liked the product they built. They couldn't see a path to making genuinely high-quality, discerning presentations that people in investment banking or consulting would rely on, and they felt the technology limited them to being a tool for individuals and students rather than professionals. They considered waiting for the models to improve, from GPT3 toward something like GPT4, but concluded the core problem wasn't raw reasoning power. The model simply lacked enough context about the presenter, the audience, and the relationship between them, and no amount of general reasoning could fix that gap.
Searching for a B2B Use Case 5:00
Rather than wait, the team asked why they had started the company in the first place, which was to help professionals tell hard, expert stories. They looked through their roughly 25 million users and found that sales and marketing teams were the clearest B2B use case. They ran twelve free pilots with mid-sized companies, initially offering to build sales decks and proposals, but customers kept asking for more, like research help, lead qualification, and account expansion support. Following that demand meant asking for access to customers' CRM, call recordings, and data warehouses, at which point the team discovered that the hardest problem wasn't generating content but making sense of messy, conflicting data spread across disparate systems.
Building and Rebuilding the Product 7:31
The team first built a go-to-market assistant that people liked but wouldn't pay for, since it relied on data owned by other systems and competed with ten similar tools. They shrank the team and spent about four months building a CRM from scratch, but then struggled to find anyone willing to use a brand-new, unproven CRM. Their solution was unconventional: they offered free desk space in their office to any startup willing to use the product, which brought in ten companies who used it daily despite its rough edges, giving hourly feedback about what was missing or broken.
Modeling the Business Instead of Automating Tasks 9:31
Keith says their naive but useful instinct was that most CRMs are built to help reps not forget things, power low-level automation, and generate forecasts, while Lightfield's team focused instead on high-fidelity modeling of the business and customer relationship. The belief was that if you could fully model a company's reality, everything else, including prompts and automated actions, would follow more easily. Three of the five founders came from Facebook, and they drew on that background to treat the customer relationship like a chronological activity log, similar to a timeline, tracking every outreach, meeting, document, and product interaction over time.
A Semi-Structured Record That Beats a Data Lake 13:00
Testing showed that a fully unstructured approach made queries too slow, so the team settled on a semi-structured design where large amounts of unstructured data live inside a structured activity log, letting the system infer cause and effect more efficiently than a plain data lake could. This lets a customer success manager ask an open question like whether an account is ready for expansion, and the system can trace through people interactions, support tickets, and product usage data to answer.
Going Schemaless to Avoid the CRM Trap 15:00
Talking with CRM consultants convinced the team that the single most consequential decision in setting up a CRM is the data model, and getting it wrong is often unrecoverable because reps won't go back and refill old records. Lightfield's answer was to be effectively schemaless: connect email, call recordings, and data warehouses first, let the system assemble relationships automatically, and let customers define or change fields later without losing historical data. Keith summarizes the underlying philosophy as intelligence being greater than schema, comparing it to old databases where every character of storage had to be rationed and predefined, a constraint that no longer applies.
Greenfield Versus Brownfield Adoption 19:00
The conversation turns to the challenge of selling into brownfield markets already dominated by incumbents like Salesforce or SAP, where the saying goes that such companies have hostages rather than customers. Lightfield initially lacked a sharp thesis for cracking brownfield accounts, so the team focused on winning new, early-stage companies first, doing direct outreach to YC startups. Over months of watching customers scale from zero reps to a hundred, they realized the real wedge wasn't automating tasks like lead scoring or outbound emails, which Salesforce could also do, but helping companies better understand themselves during the chaotic early growth period.
Overcoming Resistance From Sales Leaders 23:00
Keith raises the practical problem that greenfield companies often hire brownfield-minded VPs of sales who insist on using familiar tools like Salesforce regardless of product quality, recalling his own experience refusing to pay for Salesforce and using a free tool called Sugar CRM before eventually giving in once he hired a VP of sales who demanded it. Lightfield's response was to give the product away free to everyone in a company on sales-led plans, so engineering, finance, and customer support all rely on it, creating internal network effects that make it harder for a single new sales hire to rip out. In a few cases, a newly hired VP of sales has pushed back in favor of familiar tools, only to be overruled by the rest of the company, which had already built its workflows around Lightfield.
Simplicity over switches and knobs 26:01
The Lightfield team describes leaning into pragmatism rather than forcing a single way of working. You can still run meetings from a spreadsheet view, use dashboards and table views, or go further with a CLI if you want the frontier experience. They compare this to a car cockpit full of buttons versus a newer design with almost none, noting that the newest Teslas no longer even have a gear stalk. The point is that intuitive design, like schema-less systems, can replace a wall of controls without losing power, since plain language intelligence removes the need for many of the old knobs and switches.
Reinventing old sales workflows 28:01
Deterministic dashboards remain popular because people like checking the same view every morning, but other workflows have been fully reimagined. A classic example is the sales sequence, where you once had to map out emails, triggers, arrows, and conditions by hand. In Lightfield, you instead chat with an agent, which writes a recipe based on your world model and runs with it. Some sales leaders resisted at first, wanting their old controls back, but many came around once they saw it required less effort and worked better.
Staying open as a system of record 29:31
Lightfield sees itself as a neutral system of record, insisting the data belongs to the customer, not the company. They built their own email sync, Slack sync, and data warehouse sync, and take pride in their database's accuracy and performance. Some customers have tried building their own harness on top of Lightfield, using MCP or a CLI, but after a few weeks they usually find that Lightfield's built-in entity recognition, precision, recall, and speed outperform what they built themselves.
Finding the right pricing model 31:00
Pricing in this space is tricky because seat-based pricing no longer matches how software gets used, and outcome-based pricing is hard to define when the outcome itself is ambiguous. Lightfield started at both extremes: pure seat pricing, which customers liked but which failed because heavy users consumed ten thousand times more than light users, and pure consumption pricing, which led to a painful stretch where signed-up customers used almost nothing. They eventually split their product into four buckets: everyday CRM work covered by a flat platform fee, pipeline generation charged by consumption since it directly creates revenue opportunity, workflow automations like routing a new demo signup to the right salesperson based on company type, and intelligence and forecasting, which they see as the least discovered but most valuable part of the product. They also note that charging for outcomes is hard because results depend heavily on a customer's own product-market fit, so for now they charge for the work rather than the outcome.
Building with generalists, not swim lanes 36:30
At their earlier company Tome, teams were split into rigid functional swim lanes, which made the company slow to pivot. At Lightfield, with about 40 people, nobody owns a fixed lane; everyone attends the same daily standup, problems are stack-ranked, and whoever is free picks one up, whether it touches design, engineering, or customer success. Continuous planning means the list of priorities can change daily. This works partly because AI tools let anyone ramp up on a customer account, touch the design system in Figma, or create tasks in a linear workflow, so most people function as generalists while specialists lean toward related projects.
Speed, expansion, and staying ahead of Salesforce 40:00
The biggest worry is speed, illustrated by a story about a startup CRM that took four months to build dashboards for a client, pushing that client back to Salesforce. To avoid the same fate, Lightfield treats itself as an expansion company in a competitive market, prioritizing the accounts most likely to grow fastest over three to five years rather than just the average customer. They also note that Silicon Valley overhypes AI while much of the rest of the world underuses it, so they treat local customers partly as reference logos that make it easier to win trust in unfamiliar verticals, since buyers strongly prefer choosing a system already used by others in their own industry.
Do-it-yourself builders and future excitement 47:31
Early on, seed-stage founders sometimes tried to build their own CRM over a few weekends instead of paying for one, but usually returned once they hit hallucinations and broken emails. Larger companies rarely try to build their own system of record, but many attempt to build their own internal company brain or business world model, only to find that modeling customers accurately is much harder than expected. Looking ahead, the excitement is about Lightfield becoming a kind of crystal ball for scenario planning, helping a company decide how many salespeople to hire or what product to build next, as when one enterprise-focused customer discovered through Lightfield that it needed a new mid-market product line.
Staying focused on customers 51:01
Keith stresses that success comes from being maniacally focused on customers and building a product that truly solves their pain. He recalls that during tough transitions, employees fixated on distractions like nostalgic office vibes, food quality, or stock option repricing, but none of that actually mattered. His approach was to block out that noise entirely and keep attention on the core problem.
Closing praise for Lightfield 51:30
The hosts close by expressing deep admiration for what Keith has built, calling the product incredible and encouraging listeners to try it at lightfield.app. They thank him for the conversation and express excitement about the company's future.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
