Clarify vs. Lightfield: how to evaluate both on your own customer history

The short version
Clarify and Lightfield both put AI agents to work on CRM data. The fairest way to compare them is on your own customer history, not a demo dataset.
Lightfield is built to deliver at scale. It has predictable pricing, a white-glove rollout, and AI you can trust in production. It is an agentic CRM: it captures every customer interaction, organizes that history in a temporal context graph so agents can work from it, and runs agents inside a harness that controls what they see, do and are allowed to do.
Evaluate both on cross-account analysis, record updates, follow-up quality, workflow control, permissions, and rollout and pricing.
What Clarify says about itself
Clarify describes a full-stack AI CRM: one system for CRM, prospecting, enrichment, reporting and call intelligence. Its MCP page says Claude, ChatGPT and Codex can connect to its MCP server to retrieve, create and update records using natural language.
How Lightfield works
Lightfield is a stack, and each layer does a specific job:
- Capture and Knowledge. Every email, meeting and call is captured, and company Knowledge feeds in alongside it. You own both, along with the Skills and automations you build.
- Temporal context graph. Customer data is stored with its relationship history, so the record shows what changed, when and why. Asking about "Acme" resolves to the right record. Lightfield then reads that record and everything connected to it (contacts, meetings, opportunities, email threads) at once. Writes are proposed and held for review.
- Agent loop. For a complex question, the agent plans a response, finds and reads the relevant records, and computes with code. It then verifies the result, with every claim tied to a source, conflicts flagged, and gaps triggering another pass. The loop expands to match the question.
- Agent harness. The harness controls what the agent sees (context loading and memory), what it can do (CRM-specific tools and a code sandbox next to the data), how it is checked (your Knowledge and Skills), and what it is allowed to do (permissions, loop control and human approvals). For hard questions, it splits the work across subagents and your full history so answers stay consistent and accurate.
- Chat, automations and the product. You work with all of this through chat, automations and the UI. An automation engineer agent builds automations for you. It asks follow-up questions, tests before anything goes live, and monitors every run.
Lightfield also has a REST API, an MCP server at mcp.lightfield.app that uses OAuth 2.1, and a CLI. What each can reach depends on workspace permissions.
Run the same test on both products
Each test below targets one criterion. Run all six in both products, on your own data. Don't judge the answers on how fluent they sound. Judge whether they are correct, complete, repeatable and traceable.
1. Cross-account analysis
Ask: "Across all my open accounts, which have gone quiet after a call?" Then ask something that needs judgment: "Why did we lose our last five deals, and which open deals share those traits?"
What to look for: Does it find every matching record, or estimate? Ask the same question three times. Do the counts match each run? Can it show which records and which sources each claim came from? Lightfield's harness is built for this: it reads the full history, computes with code rather than guessing, and verifies every claim against a source before it answers.
2. Record updates
Ask: "Update this account from the last two calls and show me what changed."
What to look for: Can you see the proposed change before it applies? Can you see what the value was before, and why it changed? Does the record keep its history, or only the latest value? In Lightfield, writes through the temporal context graph are proposed and held for review, and each change keeps its history.
3. Follow-up quality
Ask: "Draft a follow-up that quotes what the buyer said on our last call."
What to look for: Is the quote accurate and from the right call? Does the draft use context from earlier meetings and emails, or only the last transcript? Does it match your team's voice and playbook? Lightfield drafts from the whole connected record and checks against your company Knowledge and Skills.
4. Workflow control and auditability
Ask: "Change this automation and show me where I approve its actions." Then: "Show me the log of everything the agent changed this week. Can I reverse a change?"
What to look for: Who builds and tests the automation? Is every run monitored? Is there a full audit log? Is there an undo for agent actions? Ask each vendor directly. Clarify's own materials state there is no undo for agent actions. Lightfield gives RevOps control over what the AI changes, with a full audit log, and risky actions wait for human approval.
5. Permissions
Ask: "Show me what a teammate without admin rights can see and change."
What to look for: Do permissions apply to agents as well as people? Does the agent see only what the person asking is allowed to see? In Lightfield, permissions apply to both, and the agent works inside them.
6. Rollout, migration and pricing
Ask: "Show me every step to move our current records in and set this up. Then show me what the bill looks like at 25 AEs, and at 100."
What to look for:
- Migration: Is there a Salesforce importer? A sandbox to test the cutover? Who owns the rollout? Lightfield offers a dedicated forward-deployed engineer, a phased Salesforce migration and a sandbox to test the cutover.
- Pricing: Does the bill follow headcount, or does it change with every recorded meeting and agent run? With Lightfield, AI usage is included. The bill follows headcount from 25 to 100 AEs, with alerts before you reach the limit. Ask Clarify how credits apply to recorded meetings and agent runs.
- Staying power: Ask how big each team is and how much each company has raised. Lightfield is about 60 people and raised a $47M Series A led by Andreessen Horowitz in September 2026. Its Head of AI, Ves Stoyanov, co-authored RoBERTa, BART and XLM-R (97,000+ citations).
Where Lightfield has been proven at scale
- Intent HQ: 130+ seats and seven-figure deals. They moved off Attio and now forecast from every customer conversation. Their CEO calls it "probably 10x the value."
- Clarium Health: They sell to health systems like Mount Sinai. They moved off HubSpot with 24 months of history and now run live board reporting in Lightfield.
Where Clarify fits
From Clarify's own description, it suits a team that wants CRM, prospecting, enrichment, reporting and call intelligence in one system, and wants to reach its CRM data from AI assistants through an MCP server. If that's the model you want, you're the buyer Clarify is describing.
FAQ
What is an agentic CRM?
A CRM built so agents can do consistent work from your customer history. Lightfield captures interactions, organizes them in a temporal context graph, and runs agents in a harness that controls what they see, do and are allowed to do.
What is Lightfield's agent harness?
It's the layer that manages the agent. It loads context by searching the graph, gives the agent CRM-specific tools and a code sandbox, checks answers against your Knowledge and Skills, and applies permissions and human approvals. For complex questions it splits the work across subagents.
What is a temporal context graph?
It's how Lightfield stores customer data. Records are connected to the people, meetings, opportunities and emails around them, and every change keeps its history, so you can see what changed, when and why.
Does Lightfield have an MCP server?
Yes. It's at mcp.lightfield.app and uses OAuth 2.1.
Does Lightfield have a public API?
Yes. Lightfield has a REST API, and API, MCP and CLI access come with every plan.
Is there a Lightfield CLI?
Yes. The docs include a CLI quickstart with Homebrew and Go install steps, and it works through the API.
Which assistants connect to Clarify's MCP server?
Clarify's page lists Claude, ChatGPT and Codex.
How does Lightfield handle security?
Lightfield is hosted on AWS. Sensitive data is encrypted at rest and in transit, and access is role-based. Lightfield is SOC 2 Type 2 audited, and you can request the full report from the Trust Center. Customers who need to store protected health information can request a Business Associate Agreement.
Next step
Connect your inbox and calls, then run the six tests above on your own data.


