Hone
Hone · Hone: Persistent AI Engines That Own Business Outcomes Such as Lead Conversion, Vendor Spend and Credit Risk
Hone builds Engines: persistent enterprise AI agents that you assign a business outcome, such as inbound lead conversion or lower vendor spend, rather than a task. An Engine connects to your systems, is tested against your past work before it acts, then works with your team in Slack, Teams and email. It is aimed at companies that want to hand an ongoing, measurable goal to AI under approval policies, and access is sales-led early access.
Best for
Mid-size and large companies with a clear, measurable business outcome (lead conversion, vendor spend, credit losses, adoption) that they want an AI system to own on an ongoing basis, with enough historical data to test it on and a team ready to give it feedback in Slack, Teams or email
Not ideal for
Individuals and small teams looking for a self-serve assistant, buyers who need published pricing or a trial before talking to sales, developers who want a documented API or SDK to build agents themselves, and organizations that must review full technical documentation before granting an AI system access to business systems
Who it's for
Business and operations leaders in revenue, procurement, risk, product and operations teams at mid-size and large companies who want to hand an ongoing, measurable outcome to AI
Capabilities
- Outcome-based setup: you give an Engine a goal, define what success looks like, set its boundaries and decide what it owns, instead of scripting individual tasks
- Self-onboarding: after you connect systems and introduce the people and processes involved, the Engine learns the company's systems, history, decision makers and constraints
- Simulation on your history: before going live, an Engine runs against past tickets, deals and decisions so you can see what it would have done, correct it and choose when it is ready
- Works in Slack, Microsoft Teams and email, where any teammate can ask, correct or hand off work, and corrections improve the Engine for everyone
- Organizational memory that stores user preferences and company best practices across people and runs, and updates as it learns
- Engine repository: each Engine writes the routines that handle its work and refines them in a versioned repository, turning repeatable work into reusable code
- Built-in evaluation: set precise metrics or let the Engine build evaluations from its own work and expert judgment, with changes replayed against past cases before they go live
- Approval policies so sensitive actions are always approved first, with every action and permission logged for audit
- Fine-grained access controls across organizations, teams, users and tools
- Orchestration of fleets of agents that fan out across tasks and collaborate with people across the company
- Example Engines for inbound lead conversion, vendor management, credit risk, product adoption and customer commitments, plus custom Engines built with Hone around your own outcome
- Hosting fully managed in Hone's cloud or inside your own VPC, with support for customer-controlled keys and network boundaries
Limitations
- Access is through a demo request while Hone rolls out early access; no self-serve signup or free trial is documented
- No pricing is published
- No public documentation, API reference or connector list is available; Slack, Microsoft Teams and email are the only work surfaces the site names
- Hone does not say which AI models power Engines
- Built as a long-running deployment: Engines onboard on company history and are pitched as improving with tenure, so they are not suited to quick evaluation or one-off tasks
- Pre-launch testing relies on simulated replays in which agents stand in for people and tools; Hone's research post notes that replays diverge further from what actually happened the longer they run, so each change is replayed several times
- Engines were introduced publicly on October 8, 2026, so there is little independent track record yet
Use cases
- Assigning an Engine to inbound lead conversion so it enriches, qualifies, routes and follows up on new leads across sales and marketing systems
- Giving an Engine responsibility for vendor spend, so it analyzes spend and license usage, gathers missing context from employees and drives consolidation and renegotiation
- Having an Engine investigate exposures across borrowers and portfolios and propose interventions to reduce credit losses
- Tasking an Engine with product adoption, so it finds friction and coordinates in-app activations, lifecycle emails and roadmap input across teams
- Tracking customer commitments from pre-sales through delivery, with the Engine chasing dependencies and handoffs between teams
- Back-testing a proposed AI process on past tickets, deals and decisions before allowing it to act on live systems
Our take
Hone is betting that the next step for enterprise AI is ownership of a metric, not more task automation, and its design follows from that. Engines onboard on company history, are tested against past decisions before acting, and replay their own changes before promoting them. These are sensible answers to the real risk of letting AI change live business processes, and approval policies, action logs and VPC hosting cover the questions security teams will ask first. What a buyer cannot do yet is check much of it independently. There are no docs, no pricing, no connector list and no model details, and Engines were announced in October 2026. Treat Hone as a design-partner engagement: pick one outcome with a clear metric and good historical data, insist on seeing the simulation results before anything goes live, and compare the effort against agents already built into your systems of record, such as Agentforce.
Who should use it
Revenue, procurement, risk, product or operations leaders at companies with a measurable outcome they would like AI to own on an ongoing basis, enough historical tickets, deals or decisions to test against, and the appetite to work closely with a young vendor during early access.
Who should skip it
Teams that want a self-serve assistant they can try today, buyers who need public pricing and documentation before engaging, developers looking for an agent framework or API, and organizations unwilling to connect business systems to a product that launched publicly in October 2026.
Strengths
- Outcome-based model: you set a goal and guardrails rather than scripting every task
- Engines are tested against your past tickets, deals and decisions before acting live, and later changes are replayed before they are promoted
- Works in Slack, Microsoft Teams and email rather than a separate app
- Approval policies for sensitive actions, logging of every action and permission, and fine-grained access controls
- Can run in Hone's cloud or inside your own VPC with customer-controlled keys
Weaknesses
- Early access through a demo request, with no self-serve option
- No published pricing
- No public docs, API, connector list or model details
- Designed to improve with tenure, so value depends on company history and ongoing team feedback
- Very new, with Engines announced in October 2026
Hone pricing
Enterprise
Custom
- Sales-led; starts with a demo request while Hone rolls out early access
- Engines built with Hone around an outcome you choose
- Hosting in Hone's cloud or inside your own VPC
Note: Hone does not publish pricing, plan tiers or usage terms.
Technical specs
Where Hone excels
Owning inbound lead conversion
The Engine enriches, qualifies, routes and follows up on leads across sales and marketing systems, and can be tested on past leads before it touches new ones.
Cutting software and vendor spend
The Engine analyzes spend and usage, gathers missing context from employees, and drives renegotiation and consolidation, with approval policies available for sensitive actions.
Reducing credit losses
The Engine investigates exposures across borrowers and portfolios, proposes interventions and tests strategies against real outcomes.
Hone vs. competitors
Hone vs. Agentforce
Agentforce is Salesforce's platform for agents grounded in Salesforce CRM and Data 360, with ready-made service, sales, commerce and supply chain agents and published pricing. Hone is not tied to one system of record: it builds an Engine around an outcome you choose, connects it to your systems, and tests it against your past work before it acts. It has no published pricing or docs.
Hone vs. Cohere North
Cohere North is an enterprise AI workspace where employees chat with company knowledge and build their own agents and automations, deployable in a VPC, on-premises or air-gapped. Hone instead delivers Engines built with Hone around a single business outcome. They run in Hone's cloud or your VPC and work with the team through Slack, Teams and email.
Hone vs. Gemini for Business
Gemini for Business (Gemini Enterprise) gives every employee a governed work agent for questions, documents, analysis and tasks, with per-seat pricing. Hone does not target individual requests: each Engine is assigned one ongoing business outcome and measured against it over time.
Hone vs. Ana by Vertice
Ana by Vertice is a negotiation agent for software purchases and renewals, built on Vertice's pricing dataset and sold with its SaaS Purchasing platform. Hone's Vendor Management Engine is one example of a broader outcome-based system: it analyzes spend and usage, gathers context from employees and drives consolidation and renegotiation. Hone's site does not describe a pricing benchmark dataset.
Frequently asked questions
What is Hone?
Hone is a San Francisco company that builds Engines: persistent AI agents for organizations. You assign an Engine an outcome, such as converting inbound leads or reducing vendor spend. It onboards into your systems, is tested against your past work, then works with your team in Slack, Microsoft Teams and email, improving from feedback and from the results of its actions.
How is an Engine different from an AI agent?
Hone draws the line this way: chatbots answer questions you prompt, agents complete tasks you assign, and Engines own outcomes you set goals and guardrails for. In practice, an Engine is meant to keep working on its goal persistently rather than per session, decide what to do and whom to involve, write and refine its own routines, and measure itself against metrics you define.
How much does Hone cost?
Hone does not publish pricing. Access starts with a demo request on hone.com, and Hone builds each Engine with the customer around the outcome it should own.
Is Hone generally available?
Not as a self-serve product. Hone introduced Engines on October 8, 2026, and its contact form says it is rolling out early access: you describe the outcome you want an Engine to own and the team follows up.
Can Hone run in our own cloud?
Hone's site says Engines can run fully managed in Hone's cloud or inside your own VPC, with support for customer-controlled keys and network boundaries. It also lists sandboxed execution, encryption in transit and at rest, configurable data retention, a commitment not to use customer data to train models, and SOC 2 certification with a trust center.
How does Hone keep an Engine from making mistakes in live systems?
Three mechanisms are described. Before going live, an Engine runs against your past work so you can review and correct what it would have done. You can set policies that require approval before sensitive actions, and every action and permission is logged. When an Engine changes its own memory, tools or code, it replays the change against past situations in simulation and only promotes it if earlier cases still pass.
Hone vs Agentforce: what is the difference?
Agentforce is Salesforce's agent platform, with ready-made agents for service, sales, commerce and other work grounded in Salesforce CRM data, and published pricing. Hone is independent of any one system of record: it builds an Engine around a business outcome you choose, connects it to your systems and tests it against your history. However, it publishes no pricing, docs or connector list.
Integrations & fit
Alternatives to consider
Salesforce
Companies whose customer data, service processes, and sales workflows already run on Salesforce and that want agents acting directly on that data across service, sales, commerce, and employee support
Paid
Cohere
Large enterprises and public-sector organizations in finance, healthcare, telecom, energy, manufacturing and government that want employees building and sharing agents and automations, but must keep data and inference in their own VPC, on-premises or air-gapped environment with per-user and per-group token limits and spend tracking down to each agent
CustomAbout Hone
Most workplace agents wait for a prompt and finish a task. Hone's Engines are designed to be assigned an outcome and keep working on it. Setting one up has four steps on Hone's site. First you give the Engine a role: define what success means, set boundaries and decide what it owns. Then it onboards: you connect your systems and introduce the people and processes it will work with, and it builds a model of how the company works, who makes decisions and what has happened before. Next comes simulation: before touching anything live, the Engine runs against your past work (old tickets, deals and decisions), you see what it would have done, correct it, and decide when it is ready. Finally it collaborates in Slack, Microsoft Teams and email, where anyone on the team can ask it questions, correct it or hand it work. Under the hood, Engines keep an organizational memory of preferences and best practices, write their own routines into a versioned repository so repeatable work becomes reusable code, and coordinate fleets of agents across tasks. Hone also describes a built-in evaluation layer: you define metrics, or the Engine builds evaluations from its own work and expert judgment, and changes to an Engine are replayed against past cases in simulation before they go live. Hone's research post explains how this works: a "director" agent mocks people and system events during a replay, and other agents stand in for tools such as Salesforce that cannot be rewound. The example Engines on the site are Inbound Conversion, Vendor Management, Credit Risk Manager, Product Adoption and Customer Commitment, and Hone says it builds a custom Engine with each customer around the outcome they choose. On governance, the site lists custom policies so that sensitive actions are always approved first, logging of every action and permission, fine-grained access controls across organizations, teams, users and tools, sandboxed execution, a commitment not to train models on customer data, SOC 2 certification with a trust center, and hosting either in Hone's cloud or inside your own VPC with customer-controlled keys. The tradeoffs come from how new and closed it is. Hone introduced Engines on October 8, 2026, access goes through a demo request while Hone rolls out early access, no pricing is published, and there are no public docs, API reference, connector list or model details. It is also a long-term deployment rather than a tool to try for an afternoon: Hone pitches Engines as getting better the longer they work, so their value depends on useful company history and ongoing feedback from the team.
Updates from Hone
In its "Introducing Engines: AI with Motion" post, Hone introduced Engines, persistent AI agents that are assigned outcomes rather than tasks, onboard themselves into a company and improve as they work. The same post says Hone raised a $60 million seed round from Benchmark and Index. Hone also published a research post on how Engines replay proposed changes in simulation before promoting them.
Are you the founder? Claim this listing →