# Framework phases

> Assess, equip, engage and improve. Every domain goes round the loop, at its own pace, for as long as agents matter to the organization.

Source: https://agentrelations.org/framework/phases

[Assess](https://agentrelations.org/framework/phases#assess)[Equip](https://agentrelations.org/framework/phases#equip)[Engage](https://agentrelations.org/framework/phases#engage)[Improve](https://agentrelations.org/framework/phases#improve)

The loop is not a project plan with an end. It is a rhythm. A new team starts with an assessment, equips the cheapest gaps first, and keeps engaging while it improves. The same loop is repeated for each domain, and several domains can be in different phases at once.

## 1. Assess

See yourself as agents see you: what they find, what they get wrong and what they fail to do.

Look at yourself the way agents do, using the same tools they use. Test with real prompts, read your own logs and score each domain against the Maturity Model. Write down what you find, including what you could not measure.

**How often:** A full baseline once, then every quarter. A light check of the core metrics every month.

| Domain | Assess in practice |
| --- | --- |
| [Discover](https://agentrelations.org/framework/domains/discover) | Test which agents reach you and which get blocked. List the sources agents use for your sector and check them. |
| [Describe](https://agentrelations.org/framework/domains/describe) | Compare your facts across the site, markup, feeds and external sources. Note every mismatch. |
| [Converse](https://agentrelations.org/framework/domains/converse) | Ask the questions customers ask, through the channels agents use. Record what comes back and how long it takes. |
| [Transact](https://agentrelations.org/framework/domains/transact) | Walk each customer task as an agent would. Note where a person must step in. |
| [Trust](https://agentrelations.org/framework/domains/trust) | Review who can reach what, what logs exist, and whether a policy is written and enforced. |
| [Measure](https://agentrelations.org/framework/domains/measure) | Establish the baseline for each core metric and note which ones you cannot yet count. |

## 2. Equip

Give agents what they need: access, facts, answers and actions, in the standards they speak.

Close the gaps found in assessment. Start with the cheapest fixes that unlock the most: access, facts and answers come before actions. Use open standards, and give each project an owner and a finish date.

**How often:** Continuous, in projects of four to twelve weeks per domain, ordered by the assessment.

| Domain | Equip in practice |
| --- | --- |
| [Discover](https://agentrelations.org/framework/domains/discover) | Fix access rules, publish the sitemap and an agent guide, correct listings and register agent-facing services. |
| [Describe](https://agentrelations.org/framework/domains/describe) | Create the fact register, add structured data, publish feeds and write policies in plain language. |
| [Converse](https://agentrelations.org/framework/domains/converse) | Stand up the question endpoint, connect live data and publish the handoff route. |
| [Transact](https://agentrelations.org/framework/domains/transact) | Expose the first actions with limits, then add checkout, changes and disputes. |
| [Trust](https://agentrelations.org/framework/domains/trust) | Write the agent policy, add scoped permissions, signature verification and logging. |
| [Measure](https://agentrelations.org/framework/domains/measure) | Classify agent traffic, add interaction logging, build the prompt panel and the monthly report. |

## 3. Engage

Answer, transact and, increasingly, reach out: run the relationship day to day.

Run the relationship. Answer questions, complete transactions, hand off to people and keep promises. As the practice matures, begin reaching out to the agents and people you want to serve, under the same rules of honesty and consent.

**How often:** Always on. Daily operations, with a weekly review of exceptions.

| Domain | Engage in practice |
| --- | --- |
| [Discover](https://agentrelations.org/framework/domains/discover) | Keep listings and registries current, and respond to blocks and crawl errors the day they occur. |
| [Describe](https://agentrelations.org/framework/domains/describe) | Update the source of truth the moment something changes, and republish everywhere. |
| [Converse](https://agentrelations.org/framework/domains/converse) | Staff the handoff route, review unanswered questions and keep answers current. |
| [Transact](https://agentrelations.org/framework/domains/transact) | Process agent-created orders and requests as you do any others, and resolve disputes. |
| [Trust](https://agentrelations.org/framework/domains/trust) | Verify agents on arrival, enforce limits and respond to abuse and incidents. |
| [Measure](https://agentrelations.org/framework/domains/measure) | Collect the data continuously, and flag anomalies when they happen. |

## 4. Improve

Measure outcomes, fix what failed and raise the level, one domain at a time.

Close the loop. Review the core metrics, find the largest cause of failed tasks, fix it and test again. Once a year, re-score maturity, drop what no longer matters and choose the next level to reach in each domain.

**How often:** A monthly review of metrics and failures. An annual re-assessment of maturity and a review of the plan.

| Domain | Improve in practice |
| --- | --- |
| [Discover](https://agentrelations.org/framework/domains/discover) | Review which agents and sources brought outcomes, and adjust access and listings. |
| [Describe](https://agentrelations.org/framework/domains/describe) | Retire stale facts, widen markup coverage and add the formats agents have begun to use. |
| [Converse](https://agentrelations.org/framework/domains/converse) | Turn the most common failed questions into better answers or new endpoints. |
| [Transact](https://agentrelations.org/framework/domains/transact) | Fix the top causes of failed actions, then extend the actions available. |
| [Trust](https://agentrelations.org/framework/domains/trust) | Review incidents and near misses, update the policy and raise the verification bar. |
| [Measure](https://agentrelations.org/framework/domains/measure) | Refine definitions only with a version change, and compare against peers in the Index. |
