a16zWhy the Next Generation of Enterprise Software Looks Nothing Like Salesforce
CHAPTERS
- 0:00 – 1:22
Founding context & what Lightfield enables in the real world
Keith frames Lightfield’s origins and ambition: modeling the revenue/customer reality of a company so AI agents can act on it. The conversation opens with a concrete, high-stakes customer story (Power) that shows how a flexible “world model” can power matching and automations beyond classic CRM use cases.
- •Founding team background and why revenue ops is a rich problem space
- •Lightfield as a “business world model” built from emails, calls, and meetings
- •Example: Power models a two-sided marketplace + pharma workflows in Lightfield
- •Automations ingest external data (FDA/clinicaltrials.gov) to enrich the model
- •Real impact: accelerating discovery of treatment options in days
- 1:22 – 3:35
From Tome’s explosive growth to the decision to hard pivot
Keith recounts building Tome, an AI presentation product that hit massive usage right as ChatGPT launched. Despite traction, the team couldn’t see a path to a product they loved or that discerning professional users would rely on, which ultimately forced a hard reset.
- •Tome: GPT-assisted presentation/page generation launched around ChatGPT
- •Explosive adoption: millions of monthly users and infrastructure constraints
- •Founders’ core issue: they didn’t love the product they were building
- •Quality ceiling: hard to serve high-end professional presentation workflows
- •Realization that the tech constrained the product’s “indispensability”
- 3:35 – 5:05
Why they didn’t ‘wait for the models to get better’ (avoiding hopium)
Alex pushes on the temptation to wait for LLMs to improve. Keith explains that the limitation wasn’t just general reasoning quality—it was missing context about presenter, audience, and relationship dynamics, which meant the core product loop wouldn’t become reliably excellent just by riding the model curve.
- •Considered shrinking the team and waiting for model improvements
- •Core gap: insufficient context about audience/presenter relationship
- •General reasoning improvements don’t fix missing situational context
- •The product risked becoming a one-shot “pie-in-the-sky” tool
- •Decision: pursue a problem where context can be captured and modeled
- 5:05 – 7:29
Finding the B2B ‘heat’: pilots with sales teams reveal the real problem
The team traced their original mission—helping professionals tell expert stories—into B2B workflows. Pilots with sales and marketing users revealed that what customers really wanted wasn’t better decks, but help making sense of fragmented go-to-market data and acting on it.
- •Analyzed user base to find strongest B2B segment: sales and marketing
- •Ran ~12 pilots with mid-sized companies, initially offering presentations
- •Customer pull expanded scope: research, lead qualification, expansion insights
- •Needed access to CRM, call recordings, data warehouse, and more
- •Breakthrough: the hardest part was reconciling conflicting/incomplete systems
- 7:29 – 9:16
Early signal of product-market pull: from assistant to CRM-first principles
Keith explains a key discontinuity: a go-to-market assistant got daily use but couldn’t command pricing because it didn’t own the underlying data. Rebuilding as a first-principles CRM felt risky—until “negative pricing” onboarding produced obsessive daily usage and relentless feedback.
- •Initial product: GTM assistant with high usage but no pricing power
- •Problem: didn’t control the system of record; crowded competitive set
- •Decision: shrink team and reimagine CRM from scratch
- •Four months in the dark → adoption hurdle: nobody wants a new CRM
- •Hack: free office desks for startups using the CRM; intense daily feedback
- 9:16 – 11:18
What’s broken in CRMs: from data-entry repositories to high-fidelity business modeling
The discussion reframes CRM from a rep-facing repository into a company’s canonical model of customer reality. Lightfield prioritizes accurate, high-fidelity modeling first—then treats emails, automation, and workflows as downstream “prompts and tool calls.”
- •Traditional CRM jobs: reminders, low-level automation, forecasting
- •Lightfield’s thesis: forecasting/understanding the business is most important
- •Manual rep entry and rigid schemas are major sources of failure
- •API/schema limitations prevent modeling reality across systems
- •Focus shifts from “do the work” to “model the company correctly”
- 11:18 – 14:56
Architecture choices: activity log as the primitive + semi-structured world model
Keith outlines the architectural bet: model the relationship chronologically first, like a Facebook timeline for company-to-company interactions. A canonical activity log stores rich unstructured data while enabling fast queries and structured CRM fields as derived outputs.
- •Founding influence from Meta: relationship modeling as the core primitive
- •Build the activity log first: messages, meetings, docs, product usage, payments
- •Fields/stages become derived updates triggered from the activity log
- •Tried fully unstructured → queries too slow (“needle in haystack”)
- •Settled on semi-structured: unstructured log + structured traversal for speed
- 14:56 – 18:48
Schema-less setup & ‘intelligence > schema’ onboarding (plus a concrete expansion example)
Lightfield aims to remove the most consequential CRM setup failure: picking the wrong data model early. By syncing core systems and assembling relationships first, customers can evolve fields later—refilling them from the activity log—while still enabling complex questions like expansion readiness.
- •CRM consultants’ biggest lever: the data model; wrong schema is fatal
- •Lightfield’s approach: connect email/call recorder/warehouse and assemble reality
- •Change fields later by re-traversing the canonical activity log
- •Example workflow: CSM asks “ready for expansion?” using tickets + usage + interactions
- •Custom objects/relationships enable unusual business models beyond standard CRM
- 18:48 – 25:27
Greenfield vs. brownfield: winning startups first, then finding the wedge to displace incumbents
They initially lacked a crisp plan to break into incumbent-heavy CRM deployments, so they started with new companies for iteration speed. Over time, they discovered the brownfield wedge isn’t “AI that sends emails,” but better steering of the company via deeper understanding and modeling.
- •Early go-to-market: target new companies (YC/startups) for faster iteration
- •Observed customers scaling rapidly (0 reps → 100 reps) while staying on Lightfield
- •Most AI-CRM messaging focused on task automation; Lightfield bets on company understanding
- •Brownfield wedge: help leaders navigate chaotic scaling with better reality modeling
- •Strategy evolves through close listening and longitudinal customer observation
- 25:27 – 30:55
Product trade-offs: skeuomorphic tables vs. natural language agents (and where each wins)
Lightfield chooses pragmatism over ideology: support familiar dashboards and spreadsheet views while also enabling agentic, natural-language workflows. Keith argues some deterministic reporting habits persist, but classic workflows like sequences can be reimagined as agent-written “recipes.”
- •Not religious about interface: tables/dashboards coexist with chat/agent UX
- •Leaders still want deterministic dashboards as a daily ritual
- •Reimagined workflow: sequences become agent-authored recipes driven by the world model
- •Change-management tension: some leaders want “knobs and switches”
- •Over time, efficiency + lower learning curve wins skeptics over
- 30:55 – 36:22
Pricing in the AI era: balancing seats, consumption, and what ‘outcomes’ really mean
They experimented with both seat-based and pure consumption pricing—and found failure modes in each. The resulting model is hybrid: fixed platform/seat fees for core CRM reliability, with consumption for value-generating work like pipeline generation, automations, and advanced intelligence.
- •Seat pricing aligned with incumbents but mismatched usage intensity (head vs tail)
- •Pure consumption caused customers to stop using the product (fear of variable cost)
- •Identified work buckets: core CRM ops vs pipeline generation vs automations vs intelligence
- •Examples of paid automation: lead routing based on research/fit
- •Outcome-based pricing is hard because results depend heavily on customer PMF
- 36:22 – 40:03
Culture and velocity: ‘everyone owns product & CS’ and continuous planning
Keith contrasts Tome’s slower, siloed “playing house” structure with Lightfield’s generalist operating system. A single daily standup, stack-ranked problems, low barrier to start projects, and a high shipping bar (bug bashes) drives rapid iteration in a fast-changing AI landscape.
- •Rejected swim lanes; cross-functional ownership to preserve pivotability
- •Everyone attends the same standup; problems are stack-ranked and reassessed continuously
- •Generalists enabled by tooling: LLMs help ramp on customers, Figma, Linear tasks
- •Low bar to begin projects; high bar to ship to customers
- •Company bug bashes ensure quality before external release
- 40:03 – 49:04
Biggest worry: speed as the moat against Salesforce gravity
Keith’s paranoia is that a startup CRM can still lose customers to Salesforce if it can’t deliver critical features—especially dashboards—fast enough. The wedge is strong for new companies, but retaining them requires out-executing incumbent expectations as customers scale.
- •Fear driven by examples of startups churning back to Salesforce
- •Dashboards and “must-have” enterprise basics are retention-critical
- •Lightfield’s wedge: best CRM for new companies—if it keeps pace with growth needs
- •Speed is framed as the primary existential risk
- •Prioritization pressure increases as customers scale and demands diversify
- 49:04 – 52:03
Looking ahead: scenario planning ‘crystal ball’ + pivot advice for founders
Keith describes the long-term vision: Lightfield as a mirror and planning engine for the most consequential decisions—hiring, product direction, market focus—powered by frontier intelligence over a unified company model. He closes with hard-won pivot guidance: ignore noise, find real pain, and obsess over customers.
- •Future focus: scenario planning (reps to hire, what to build next, where to go)
- •Customer example: discovering the need for a mid-market product line via Lightfield
- •Silicon Valley customers as reference logos to cross the chasm into broader markets
- •DIY is overhyped: building a real business world model is harder than it seems
- •Pivot advice: block out distractions, find pain, and stay maniacally customer-focused
