All Insights

Articles

What Is AI Agent Assist? AI Assistants for Call Centre Agents Explained

AI agent assist is a real-time assistant for human call centre agents: how it works, what a 5,172-agent study found, and when a voice agent fits.

SUMMARIZE WITH:
SHARE THIS POST

Last updated: 9 October 2026. Documented research, not a hands-on evaluation: no agent-assist product was trialled. Sources were read on 9 October 2026.

AI agent assist, or an AI assistant for call centre agents, is software that helps a human agent on a live call. It transcribes the conversation, surfaces answers and next steps, and drafts the notes afterwards.

The best evidence on agent assist is one large field study of 5,172 support agents. Productivity rose 15% on average and 30% for less experienced and lower-skilled agents, with small gains in speed and small declines in quality for the most skilled.

"AI assistant" means two different things in a call centre

Vendors use "AI call centre assistant" for two products that do opposite things:

  • Agent assist (agent-facing). The assistant helps a human who stays on the call. The caller only ever talks to the person. This page is about this one.

  • An AI voice agent (customer-facing). The AI answers the call itself and talks to the caller. It resolves the request, or hands over to a person when it can't.

One improves the human on the call; the other removes the call. Brilo makes the second kind and does not make agent assist, so this page explains agent assist on its own terms and says when it is the better choice.

How agent assist works during a call

Products describe broadly the same flow:

  1. Before the call: as a call comes in, the assistant shows the agent a summary of the customer's previous contact.

  2. During the call: it transcribes the conversation live, as both people speak.

  3. From the transcript, it works out what the customer wants and surfaces help on screen: the relevant knowledge article, a suggested reply, a compliance prompt or a next-best action.

  4. The agent decides. They accept, edit or ignore each suggestion. Nothing is said to the customer unless the person says it.

  5. After the call: it drafts the summary, disposition and notes, and can update the CRM.

Everything depends on the transcript: a misheard account number produces a confident wrong suggestion. See AI call transcription for how it is measured and priced.

The parts of an AI assistant for agents

Component

What it does

When

What it depends on

Live transcription

Turns the call into text as it happens

During

Audio quality, accents, vocabulary

Knowledge answers

Finds the right article or answer from your own content

During

An accurate, current knowledge base

Suggested replies

Proposes what to say next

During

Good examples from your best agents

Next-best action

Recommends an action: escalate, retain, book, refund

During

Clear rules and past outcomes

Compliance prompts

Reminds the agent of required wording or checks

During

Your written policies

Live sentiment cues

Flags a caller who is getting frustrated

During

Transcript and, in some products, voice tone

After-call work

Drafts the summary, disposition and CRM notes

After

The transcript

After-call work is the least discussed part and often the one agents notice most, because it replaces typing.

What the 5,172-agent study found

This category has a rare thing: a large field study. Brynjolfsson, Li and Raymond studied the staggered rollout of a generative AI assistant that suggested replies to 5,172 customer-support agents at one firm. The published version (Quarterly Journal of Economics, 2025) reports:

  • Issues resolved per hour rose 15% on average.

  • Less experienced and lower-skilled agents resolved 30% more issues per hour, improving both speed and quality.

  • The most experienced and highest-skilled agents saw small gains in speed and small declines in quality.

The earlier 2023 NBER working paper version (w31161) reports further findings:

  • New agents caught up faster. Treated agents with two months of tenure performed as well as untreated agents with more than six months.

  • Average chat duration fell 3.8 minutes, a 9% decline from the baseline.

  • Customers treated agents better, measured by the sentiment of their messages, and were less likely to ask for a supervisor.

  • Attrition fell, "driven by the retention of newer workers".

The limits belong with the numbers: text chat, not voice, at one firm, with 2023-era tooling. "15% productivity" is an average across a population that may look nothing like your team.

Who agent assist helps least

The gains concentrated at the bottom of the experience curve. For the most experienced and highest-skilled agents, the published version reports small declines in quality. The tool spreads the practices of better agents, which helps those who lack them. That decides the purchase more than the vendor does:

  • A large, new or high-turnover team is where the measured 30% sits: agent assist shortens the ramp.

  • A small, tenured, expert team was measured at small gains in speed and small declines in quality.

  • Adherence matters. In the 2023 working paper version, agents who followed the recommendations more closely gained more, and adherence rose over time.

We found no independent study of agent assist on voice calls, or with tools newer than 2023.

Why human agents are staying, and getting harder calls

Service leaders are adding to agent roles, not removing them (Gartner surveys and predictions):

  • In a survey of 321 service leaders (September–October 2025), 85% are adding tasks to agent roles, 75% are shifting agents into new roles, and 63% are reducing headcount gradually through attrition (Gartner, 28 Apr 2026).

  • In a separate survey of 5,801 US customers, 54% trust human agents more than AI for recommendations, against 32% who trust AI more (same release).

  • 58% of leaders aim to upskill agents into knowledge-management specialists, and 84% plan to add new skills (Gartner, 18 Feb 2026).

  • Prediction: by 2028, "regulatory changes related to AI will increase assisted service volume by 30%" (Gartner, 26 Jan 2026).

Once AI takes the routine calls, those left for people are harder. For the jobs evidence, see will AI replace call centre jobs?.

Most call centres need both an assistant for their people and an AI that takes the routine calls. Brilo is the second half. Its Free plan includes 10 minutes a month for testing. See Brilo pricing.

Agent assist vs an AI voice agent


Agent assist

AI voice agent

Who talks to the caller

Your human agent

The AI

What changes

The agent gets faster and more consistent

Routine calls no longer reach a person

Best calls

Complex, emotional, judgement-heavy

Routine, repetitive, after-hours, overflow

Best evidence

One field study: +15% average, +30% less-skilled agents (chat; QJE 2025)

No independent containment figures published, by Brilo or anyone else we checked

Typical pricing

Per agent seat per month, where published

Per minute or bundled minutes

Headcount effect

None by itself

Fewer calls for people; whether staffing changes is your decision

What it can't do

Help with a call nobody picks up

Judgement it wasn't given; calls that need a person

Agent assist cannot answer a call at 2am; a voice agent cannot comfort a caller whose complaint has gone wrong.

When agent assist is the right answer

Buy agent assist, not what Brilo sells, when:

  • Your calls need judgement. In complex or multi-issue conversations, a voice agent would escalate most calls, and you would pay for an agent that mostly transfers.

  • Your calls are emotionally loaded. Bereavement, medical anxiety, and complaints that have already gone wrong. A person should take these, and helping that person is the useful intervention.

  • A human must be accountable. Some regulated conversations need a person to give the advice. Real-time compliance prompts are the right shape of help.

  • You are not going to change headcount. Then a voice agent's saving is theoretical, and making the people you keep faster is the real return.

  • You have high turnover or many new agents, where the measured gains sit.

When an AI voice agent is the right answer, and why most need both

A voice agent fits when:

  • The calls are routine and repetitive: opening hours, order status, booking, triage, qualification. Volume is high, judgement is low, and a person adds availability, not much else.

  • The calls currently go unanswered: after hours and overflow. Agent assist cannot help with a call nobody picks up.

  • Volume is spiky. Concurrent capacity scales in a way rosters don't.

  • The constraint is cost per call, not quality per call.

For a contact centre of any size, the realistic answer is to route by call type. The AI voice agent takes the routine and out-of-hours volume. People take the complex and emotional calls, with agent assist helping them.

The catch: once routine calls are gone, the remaining mix is harder, so handling time rises. Budget for more supervision, not less. For the capacity side, see AI workforce management and diallers.

What agent assist costs, where it's published

Vendor-published list prices, checked 9 October 2026; not performance claims.

Product

Published list price

Zendesk Copilot (add-on)

$50 per agent per month, paid yearly, on top of a Zendesk plan (zendesk.com/pricing)

CloudTalk AI Conversation Intelligence (add-on)

€9 per user per month, billed annually (cloudtalk.io/pricing)

Cresta

Not published: the /pricing path returned 404

Balto

Not published: the /pricing path returned 404

Level AI

Not published: the /pricing path returned 404

Observe.AI

Not published: the /pricing path returned 404

Public prices exist only as platform seat add-ons; specialists expect a quote. The software price is rarely the decision. A ten-agent team on Zendesk Copilot pays $500 a month for the add-on and still employs ten agents. The same team's 2,000 monthly calls at four minutes each come to 8,000 minutes. On Brilo's Growth plan ($499 for 2,500 minutes, then $0.14 a minute, per the pricing page), that is about $1,270 a month. Both are small next to ten salaries. The real question is whether calls can be removed from people at all. For full cost modelling, see AI call centre costs, pricing and ROI.

What Brilo does for the people on your team, and what it doesn't

Brilo does not provide agent assist: no real-time guidance on a live call, no next-best-action prompts, no after-call work on your agents' calls. If your calls need a helped person, Brilo is not the product.

What it does offer the people who work alongside it (docs read 9 Oct 2026):

  • Warm transfers with a briefing. When the AI passes a call to a colleague, it can play a contextual summary first, so "your colleague picks up already knowing who is on the line and what they want" (Pass the call to a person).

  • Summaries written for whoever acts next, which can be sent to your team by text or email as each call ends (Personalised call summaries; Send a text or email).

  • Team access at no extra cost per person, to listen to calls and read reports (Invite your team).

  • Read-only access for AI tools. An assistant such as Cursor can answer questions about your calls through Brilo's MCP server. It never returns recordings, and HIPAA workspaces get no transcripts (Brilo MCP server).

Brilo does not publish a containment rate; ask for one on call types like yours.

Common misconceptions about AI agent assist

"It makes every agent 15% faster." The 15% was an average. Novices gained most, and experts gained little.

"An assistant replaces training." It spreads good practice faster. It still needs a knowledge base someone maintains, which is why 58% of leaders in Gartner's survey plan to upskill agents into knowledge management.

"Agent assist and voice agents are rivals." They are designed for different calls, and most operations route between them.

"Real-time means right." Suggestions inherit transcription errors. Review assisted calls as you would any others; see AI call centre quality assurance. For how both fit the wider stack, see the AI call centre guide.

AI agent assist: frequently asked questions

What is AI agent assist? Software that supports a human agent on a live call with transcription, knowledge answers, suggested replies, next-best actions and drafted notes. The human stays on the call.

Is an AI assistant for call centre agents the same as an AI voice agent? No. An agent assistant helps a person who talks to the caller; an AI voice agent talks to the caller itself. Most contact centres split by call type.

Does agent assist actually improve productivity? In the best study (5,172 agents, one firm, text chat; Quarterly Journal of Economics, 2025), issues resolved per hour rose 15% on average and 30% for less experienced and lower-skilled agents, while the most skilled saw small gains in speed and small declines in quality.

What are the best real-time agent assist tools for call centres? We don't rank vendors, because Brilo competes with the category. Shortlist on four checks: - evidence from teams like yours - transcription accuracy on your own calls - whether it reads your knowledge base - a published price or a firm quote

For appointment-setting lines, also ask whether the bookings could go to an AI voice agent instead. A Brilo agent can book into Google Calendar during the call (Book appointments).

How much does AI agent assist cost? Rarely published. Zendesk lists Copilot at $50 per agent per month (paid yearly) on top of a plan; CloudTalk lists €9 per user per month. Cresta, Balto, Level AI and Observe.AI publish none (9 Oct 2026).

Does Brilo offer agent assist? No. Brilo is an AI voice agent. It briefs colleagues on warm transfers and sends summaries, but gives no live guidance on your agents' calls.

If routine and after-hours calls are the part you want off your team's desks, test an AI voice agent on the Free plan: see Brilo pricing.

All Insights

Articles

What Is AI Agent Assist? AI Assistants for Call Centre Agents Explained

AI agent assist is a real-time assistant for human call centre agents: how it works, what a 5,172-agent study found, and when a voice agent fits.

SUMMARIZE WITH:
SHARE THIS POST

Last updated: 9 October 2026. Documented research, not a hands-on evaluation: no agent-assist product was trialled. Sources were read on 9 October 2026.

AI agent assist, or an AI assistant for call centre agents, is software that helps a human agent on a live call. It transcribes the conversation, surfaces answers and next steps, and drafts the notes afterwards.

The best evidence on agent assist is one large field study of 5,172 support agents. Productivity rose 15% on average and 30% for less experienced and lower-skilled agents, with small gains in speed and small declines in quality for the most skilled.

"AI assistant" means two different things in a call centre

Vendors use "AI call centre assistant" for two products that do opposite things:

  • Agent assist (agent-facing). The assistant helps a human who stays on the call. The caller only ever talks to the person. This page is about this one.

  • An AI voice agent (customer-facing). The AI answers the call itself and talks to the caller. It resolves the request, or hands over to a person when it can't.

One improves the human on the call; the other removes the call. Brilo makes the second kind and does not make agent assist, so this page explains agent assist on its own terms and says when it is the better choice.

How agent assist works during a call

Products describe broadly the same flow:

  1. Before the call: as a call comes in, the assistant shows the agent a summary of the customer's previous contact.

  2. During the call: it transcribes the conversation live, as both people speak.

  3. From the transcript, it works out what the customer wants and surfaces help on screen: the relevant knowledge article, a suggested reply, a compliance prompt or a next-best action.

  4. The agent decides. They accept, edit or ignore each suggestion. Nothing is said to the customer unless the person says it.

  5. After the call: it drafts the summary, disposition and notes, and can update the CRM.

Everything depends on the transcript: a misheard account number produces a confident wrong suggestion. See AI call transcription for how it is measured and priced.

The parts of an AI assistant for agents

Component

What it does

When

What it depends on

Live transcription

Turns the call into text as it happens

During

Audio quality, accents, vocabulary

Knowledge answers

Finds the right article or answer from your own content

During

An accurate, current knowledge base

Suggested replies

Proposes what to say next

During

Good examples from your best agents

Next-best action

Recommends an action: escalate, retain, book, refund

During

Clear rules and past outcomes

Compliance prompts

Reminds the agent of required wording or checks

During

Your written policies

Live sentiment cues

Flags a caller who is getting frustrated

During

Transcript and, in some products, voice tone

After-call work

Drafts the summary, disposition and CRM notes

After

The transcript

After-call work is the least discussed part and often the one agents notice most, because it replaces typing.

What the 5,172-agent study found

This category has a rare thing: a large field study. Brynjolfsson, Li and Raymond studied the staggered rollout of a generative AI assistant that suggested replies to 5,172 customer-support agents at one firm. The published version (Quarterly Journal of Economics, 2025) reports:

  • Issues resolved per hour rose 15% on average.

  • Less experienced and lower-skilled agents resolved 30% more issues per hour, improving both speed and quality.

  • The most experienced and highest-skilled agents saw small gains in speed and small declines in quality.

The earlier 2023 NBER working paper version (w31161) reports further findings:

  • New agents caught up faster. Treated agents with two months of tenure performed as well as untreated agents with more than six months.

  • Average chat duration fell 3.8 minutes, a 9% decline from the baseline.

  • Customers treated agents better, measured by the sentiment of their messages, and were less likely to ask for a supervisor.

  • Attrition fell, "driven by the retention of newer workers".

The limits belong with the numbers: text chat, not voice, at one firm, with 2023-era tooling. "15% productivity" is an average across a population that may look nothing like your team.

Who agent assist helps least

The gains concentrated at the bottom of the experience curve. For the most experienced and highest-skilled agents, the published version reports small declines in quality. The tool spreads the practices of better agents, which helps those who lack them. That decides the purchase more than the vendor does:

  • A large, new or high-turnover team is where the measured 30% sits: agent assist shortens the ramp.

  • A small, tenured, expert team was measured at small gains in speed and small declines in quality.

  • Adherence matters. In the 2023 working paper version, agents who followed the recommendations more closely gained more, and adherence rose over time.

We found no independent study of agent assist on voice calls, or with tools newer than 2023.

Why human agents are staying, and getting harder calls

Service leaders are adding to agent roles, not removing them (Gartner surveys and predictions):

  • In a survey of 321 service leaders (September–October 2025), 85% are adding tasks to agent roles, 75% are shifting agents into new roles, and 63% are reducing headcount gradually through attrition (Gartner, 28 Apr 2026).

  • In a separate survey of 5,801 US customers, 54% trust human agents more than AI for recommendations, against 32% who trust AI more (same release).

  • 58% of leaders aim to upskill agents into knowledge-management specialists, and 84% plan to add new skills (Gartner, 18 Feb 2026).

  • Prediction: by 2028, "regulatory changes related to AI will increase assisted service volume by 30%" (Gartner, 26 Jan 2026).

Once AI takes the routine calls, those left for people are harder. For the jobs evidence, see will AI replace call centre jobs?.

Most call centres need both an assistant for their people and an AI that takes the routine calls. Brilo is the second half. Its Free plan includes 10 minutes a month for testing. See Brilo pricing.

Agent assist vs an AI voice agent


Agent assist

AI voice agent

Who talks to the caller

Your human agent

The AI

What changes

The agent gets faster and more consistent

Routine calls no longer reach a person

Best calls

Complex, emotional, judgement-heavy

Routine, repetitive, after-hours, overflow

Best evidence

One field study: +15% average, +30% less-skilled agents (chat; QJE 2025)

No independent containment figures published, by Brilo or anyone else we checked

Typical pricing

Per agent seat per month, where published

Per minute or bundled minutes

Headcount effect

None by itself

Fewer calls for people; whether staffing changes is your decision

What it can't do

Help with a call nobody picks up

Judgement it wasn't given; calls that need a person

Agent assist cannot answer a call at 2am; a voice agent cannot comfort a caller whose complaint has gone wrong.

When agent assist is the right answer

Buy agent assist, not what Brilo sells, when:

  • Your calls need judgement. In complex or multi-issue conversations, a voice agent would escalate most calls, and you would pay for an agent that mostly transfers.

  • Your calls are emotionally loaded. Bereavement, medical anxiety, and complaints that have already gone wrong. A person should take these, and helping that person is the useful intervention.

  • A human must be accountable. Some regulated conversations need a person to give the advice. Real-time compliance prompts are the right shape of help.

  • You are not going to change headcount. Then a voice agent's saving is theoretical, and making the people you keep faster is the real return.

  • You have high turnover or many new agents, where the measured gains sit.

When an AI voice agent is the right answer, and why most need both

A voice agent fits when:

  • The calls are routine and repetitive: opening hours, order status, booking, triage, qualification. Volume is high, judgement is low, and a person adds availability, not much else.

  • The calls currently go unanswered: after hours and overflow. Agent assist cannot help with a call nobody picks up.

  • Volume is spiky. Concurrent capacity scales in a way rosters don't.

  • The constraint is cost per call, not quality per call.

For a contact centre of any size, the realistic answer is to route by call type. The AI voice agent takes the routine and out-of-hours volume. People take the complex and emotional calls, with agent assist helping them.

The catch: once routine calls are gone, the remaining mix is harder, so handling time rises. Budget for more supervision, not less. For the capacity side, see AI workforce management and diallers.

What agent assist costs, where it's published

Vendor-published list prices, checked 9 October 2026; not performance claims.

Product

Published list price

Zendesk Copilot (add-on)

$50 per agent per month, paid yearly, on top of a Zendesk plan (zendesk.com/pricing)

CloudTalk AI Conversation Intelligence (add-on)

€9 per user per month, billed annually (cloudtalk.io/pricing)

Cresta

Not published: the /pricing path returned 404

Balto

Not published: the /pricing path returned 404

Level AI

Not published: the /pricing path returned 404

Observe.AI

Not published: the /pricing path returned 404

Public prices exist only as platform seat add-ons; specialists expect a quote. The software price is rarely the decision. A ten-agent team on Zendesk Copilot pays $500 a month for the add-on and still employs ten agents. The same team's 2,000 monthly calls at four minutes each come to 8,000 minutes. On Brilo's Growth plan ($499 for 2,500 minutes, then $0.14 a minute, per the pricing page), that is about $1,270 a month. Both are small next to ten salaries. The real question is whether calls can be removed from people at all. For full cost modelling, see AI call centre costs, pricing and ROI.

What Brilo does for the people on your team, and what it doesn't

Brilo does not provide agent assist: no real-time guidance on a live call, no next-best-action prompts, no after-call work on your agents' calls. If your calls need a helped person, Brilo is not the product.

What it does offer the people who work alongside it (docs read 9 Oct 2026):

  • Warm transfers with a briefing. When the AI passes a call to a colleague, it can play a contextual summary first, so "your colleague picks up already knowing who is on the line and what they want" (Pass the call to a person).

  • Summaries written for whoever acts next, which can be sent to your team by text or email as each call ends (Personalised call summaries; Send a text or email).

  • Team access at no extra cost per person, to listen to calls and read reports (Invite your team).

  • Read-only access for AI tools. An assistant such as Cursor can answer questions about your calls through Brilo's MCP server. It never returns recordings, and HIPAA workspaces get no transcripts (Brilo MCP server).

Brilo does not publish a containment rate; ask for one on call types like yours.

Common misconceptions about AI agent assist

"It makes every agent 15% faster." The 15% was an average. Novices gained most, and experts gained little.

"An assistant replaces training." It spreads good practice faster. It still needs a knowledge base someone maintains, which is why 58% of leaders in Gartner's survey plan to upskill agents into knowledge management.

"Agent assist and voice agents are rivals." They are designed for different calls, and most operations route between them.

"Real-time means right." Suggestions inherit transcription errors. Review assisted calls as you would any others; see AI call centre quality assurance. For how both fit the wider stack, see the AI call centre guide.

AI agent assist: frequently asked questions

What is AI agent assist? Software that supports a human agent on a live call with transcription, knowledge answers, suggested replies, next-best actions and drafted notes. The human stays on the call.

Is an AI assistant for call centre agents the same as an AI voice agent? No. An agent assistant helps a person who talks to the caller; an AI voice agent talks to the caller itself. Most contact centres split by call type.

Does agent assist actually improve productivity? In the best study (5,172 agents, one firm, text chat; Quarterly Journal of Economics, 2025), issues resolved per hour rose 15% on average and 30% for less experienced and lower-skilled agents, while the most skilled saw small gains in speed and small declines in quality.

What are the best real-time agent assist tools for call centres? We don't rank vendors, because Brilo competes with the category. Shortlist on four checks: - evidence from teams like yours - transcription accuracy on your own calls - whether it reads your knowledge base - a published price or a firm quote

For appointment-setting lines, also ask whether the bookings could go to an AI voice agent instead. A Brilo agent can book into Google Calendar during the call (Book appointments).

How much does AI agent assist cost? Rarely published. Zendesk lists Copilot at $50 per agent per month (paid yearly) on top of a plan; CloudTalk lists €9 per user per month. Cresta, Balto, Level AI and Observe.AI publish none (9 Oct 2026).

Does Brilo offer agent assist? No. Brilo is an AI voice agent. It briefs colleagues on warm transfers and sends summaries, but gives no live guidance on your agents' calls.

If routine and after-hours calls are the part you want off your team's desks, test an AI voice agent on the Free plan: see Brilo pricing.

Automate your business with AI phone Agents

Automate your business with AI phone Agents

Automate your business with AI phone Agents

Automate your business with AI phone Agents

Call automation for healthcare, real estate, logistics, financial services & small businesses.

Call automation for healthcare, real estate, logistics, financial services & small businesses.