← Back to all work

AI Agent Partner

From a helper panel to a conversational partner

Agent Helper helped agents read a case. AI Agent Partner is conversational: it helps them decide and act, from cited replies to escalating in Jira, swarming in Slack and scheduled follow-ups.

Role
Product owner
Scope
Conversational copilot, Jira and Slack swarming, follow-ups, reporting

Why we built it

Agent Helper made the case easier to read. But agents still left the case to get help, chase engineering in Jira, pull experts into Slack and remember follow-ups. AI Agent Partner set out to help them decide what to do next and then do it, in one conversation.

Why a partner, not just a helper

Agent Helper answered "what is this case about?" The next question agents kept asking was "what do I do now, and who do I need?" That needed a conversation with the case, connected to the tools where support work actually happens.

Agent Helper AI Agent Partner
How agents use itA panel of summaries, cards and related contentA conversation: ask in plain language, get answers and actions
Main jobRead the case faster and draft a replyDecide the next step and carry it out
ContextCase summary, timeline and related contentThe case timeline plus connected tools such as Jira and Slack
Getting helpRelated experts listed in the panelEscalate in Jira or swarm via Slack without leaving the case
Follow-throughAgents track follow-ups themselvesSchedule follow-ups from the conversation
EvidenceRelated articles and casesCited sources beside each suggestion
Where it's headingAssistive panelAn autonomous agent for basic L1 cases

See the Agent Helper case study →

Customer story: Revenera

Customer outcome · Revenera's Agent Helper deployment

Revenera's technical support team handles complex cases such as licensing disputes, installation failures and compliance audits. Instead of adding headcount, they embedded Agent Helper in their support console. In our AI Agent Partner webinar, Revenera's Director of Technical Support shared that impact and what agents still need as cases get more complex, which is the gap AI Agent Partner is built for.

55%reduction in resolution time, sustained across the period
45%of all cases resolved with Agent Helper, up from 1 in 5 at launch
2x+increase in AI-assisted cases

Revenera's reported results with Agent Helper, from SearchUnify's customer case study. They are not AI Agent Partner metrics. Read the Revenera case study (opens in a new tab)

What I owned

I owned the product across case assessment, response assistance, recommended next actions, and adoption and feedback reporting.

Case assessment

A read of where the case stands: what the customer asked, what has been tried, and what is still open.

Response assistance

Suggested replies grounded in the case and in cited knowledge sources the agent can open before sending.

Recommended next actions

One clear next step, such as requesting logs or escalating, with the reason it was suggested.

Adoption and feedback reporting

Reporting that shows managers and our team which suggestions agents use, edit or reject.

Feature list

  • Conversational

    Agents ask about the case in plain language and get answers grounded in its timeline.

  • Case assessment

    Where the case stands: what the customer asked, what has been tried and what is still open.

  • Cited response assistance

    Suggested replies with sources the agent can open before sending.

  • Recommended next actions

    One clear next step, with the reason it was suggested.

  • Jira integration

    Linked Jira issues and their status stay visible in the conversation, so agents know where engineering stands.

  • Escalate in Jira

    Raise an engineering escalation in Jira from the conversation, with the case summary and logs attached.

  • Swarm via Slack

    Pull the right experts into Slack with the case context already shared.

  • Scheduled follow-ups

    Set a follow-up from the conversation so nothing slips.

  • Adoption and feedback reporting

    Shows which suggestions and actions agents use, edit or reject.

See it in a conversation

Illustrative exampleCase 61204 · Webhook deliveries failing · fictional

Try a prompt

What's blocking this case?

The customer gets a 401 on every webhook retry since Friday. That usually means the signing secret was rotated [1]. Logs are attached, but no fix has been offered yet.

[1] KB: Webhook authentication errors

Timeline as context, sources as proof

The case timeline is the copilot's context. Every customer message, agent reply, status change and sentiment shift is ordered in time, so the copilot reasons over what happened in the case instead of the last message alone.

Each suggested reply lists its sources. Agents told us they would only send a draft they could check, so citations sit next to the sentence they support.

Case timeline (context)

  1. Customer Webhook deliveries failing since Friday
  2. Agent Asked for endpoint and error codes
  3. Customer 401 on every retry. Escalating if not fixed today.
  4. Status: Waiting on support · sentiment falling

Copilot suggestion

A 401 on every retry usually means the signing secret was rotated[1]. Could you regenerate the secret and update it on your endpoint[2]?
  1. [1] KB: Webhook authentication errors
  2. [2] Similar resolved case, same error code

Next action: Request the delivery log ID, because the error is consistent across retries.

Illustrative example Fictional case and sources.

Decisions I owned

Ground suggestions in the timeline

I prioritized timeline context over a larger prompt, so the copilot sees order and change, not a pile of text.

Cite, then suggest

A reply without a source the agent can open is treated as lower confidence and labeled that way.

Measure use, not just output

I defined adoption around what agents did with a suggestion, so the team could tell useful help from noise.

How I measured it

MeasureQuestion it answers
Draft useHow often do agents start from a suggested reply?
Edit depthHow much do they change before sending?
Source checksDo agents open cited sources, and on which kinds of cases?
Next-action acceptanceWhich recommended actions get taken, and which get dismissed?
Escalations, swarms and follow-upsHow often do agents escalate in Jira, swarm in Slack or schedule follow-ups from the conversation?
Agent feedbackWhat do thumbs-down reasons say about missing context?

Where it's heading

  1. Today

    Suggests

    Case assessment, cited replies and recommended next actions.

  2. Today

    Acts when asked

    Escalates in Jira, swarms in Slack and schedules follow-ups on the agent's request.

  3. Future

    Resolves basic L1 cases

    Runs as an autonomous agent for simple, well-understood L1 cases.

The autonomous L1 agent is a future direction, not a shipped capability.

Watch and read