Part 2
Workflows in Focus
One of the biggest reasons AI agent deployments fail is that teams try to drop them into workflows that aren't clearly scoped, structured, or suited for agentic work.
That's why this section zeroes in on where GTM teams actually need leverage. So we start with asking a more pointed question:
Where in my workflow can an agent meaningfully reduce friction, save time, or improve output?
We've mapped four real-world workflows where GTM teams are already putting agents to work:
Prospecting & Lead Management
Sales, RevOps, Alliances
Data enrichment, scoring, account research, and list hygiene
Campaign & Content Activation
Marketing, CX, Product Marketing
Personalization, copy generation, segmentation, asset triggering
Handoffs & Internal Coordination
Sales → CX, Marketing → Sales
Workflow transitions, internal notes, CRM logging, follow-ups
Post-Sale Follow-through & QA
CX, Customer Success, RevOps
Adoption tracking, renewal prep, sentiment monitoring, QBRs
Each of these workflows has its own structure, pressure points, and agent-fit profile. In this section, we'll:
- Outline what a well-scoped agent can — and can't — do
- Highlight where things go wrong when teams move too fast
- Show how leaders are stress-testing agentic adoption in the real world
Workflow 1
Prospecting & Lead Management
Relevant teams: Sales, RevOps, Marketing (ABM), Partnerships
Before any call gets booked or campaign goes live, someone has to do the grunt work. Finding accounts. Enriching contacts. Logging them into your CRM. Prioritizing based on fuzzy rules. Prepping for outreach.
It's essential work, but it's also repetitive, structured, and time-consuming. And that's exactly the kind of workflow where a well-scoped agent can thrive.
What a good agent should be able to do
A well-scoped agent in this workflow isn't replacing your SDR. It's giving them more time to do the work that moves deals forward. That means handling the repetitive, rules-based, and data-heavy tasks that slow teams down.
Here's what that requires under the hood:
Agent Skill Stack
- Data EnrichmentPulls missing contact or account info from trusted sources
- CRM HygieneAuto-detects duplicates, updates fields, logs actions
- Scoring & TaggingSorts and prioritizes accounts using ICP logic
- Account ResearchCompiles relevant public info for SDRs or AEs
- Sequencing SupportSuggests outreach steps or schedules follow-ups
What not to delegate to the agent
These tasks might seem automatable, but they still need human context, judgment, or nuance.
- Lead qualification based on tone or intent"Did they sound interested?" is still a human read.
- Account prioritization without clear scoring logicAgents follow rules. If the rules are fuzzy, so is the output.
- Relationship-based outreachKnowing when to nudge, escalate, or hold off requires social and strategic context an agent doesn't have.
- Exception handlingIf the outcome depends on emotional nuance, unstated context, or improvisation, it's not a good fit for delegation.
Big takeaway: If the outcome depends on emotional nuance, unstated context, or improvisation, it’s not a good fit for delegation.
Risks GTM Teams Flagged
The biggest risk GTM leaders flagged is trying to automate a messy process or one that doesn't exist yet.
Teams often jump to agents before they've mapped how work actually happens. Without a clear definition of what qualifies a lead, what "clean" data looks like, or where the handoff happens, even the best agents will drift or default to flawed assumptions.
This is especially true for really messy and scaling teams. They often lack stable, repeatable workflows that makes them especially vulnerable to over-automating too soon. As Seth Nesbitt put it:
Turning over our unique value proposition, how we treat leads and prospects — that's awfully core and strategic for us. I'm going to hesitate to outsource that to an AI agent, especially when our process is still being refined. (If) we don't necessarily have a fully defined process to go and automate yet… What am I going to turn to AI to automate? Something that's not set up?
And even when a process does exist, agents aren't a shortcut to understanding your own workflows. You still need to know which tasks matter for your team.
You really gotta understand the activities of your team… Taking in an AI agent won't solve anything unless you know what's going to move the needle there.
That's why framing agents as "replacements" can backfire. The job doesn't disappear; it just changes shape. And someone still needs to own the outcome.
Agents will increase your capacity, not do your job for you. If you could manage five reps, maybe now you can manage eight or twelve. But the job doesn't go away.
Big takeaway: If the outcome depends on emotional nuance, unstated context, or improvisation, it’s not a good fit for delegation.
Workflow 2
Campaign Orchestration & Personalization
Relevant teams: Marketing, Growth, Revenue Operations, CX
This is where things get messy. And expensive. GTM teams juggle multiple campaign variants, channels, and segments, but personalization often collapses under the weight of that complexity. Campaign reporting is fragmented. Engagement is spotty. And every team has felt the sting of launching a big-budget campaign that underperformed.
This is the campaign orchestration bottleneck, and a well-designed agent can help.
What a good agent should be able to do
AI agents here operate as orchestration assistants. They're not crafting the strategy, but they are executing on it. They should help map personas into segments, variants into outputs, and playbooks into live campaigns.
Agent Skill Stack
- Audience segmentationCreate dynamic segments based on behavior, firmographics, or triggers
- Variant generationSpin out campaign versions for different ICPs
- Asset coordinationMatch copy, design, and CTAs to the right audiences
- Channel orchestrationSequence outreach across email, ads, chat, and more
- Trigger-based automationLaunch workflows based on event or signal
- Campaign performance monitoringTrack micro-metrics in real-time and auto-pause or optimize
What not to delegate to the agent
As agents can handle a lot. But in campaign planning and content workflows, there are certain responsibilities you should keep human by default.
- Messaging & positioning strategyAgents can adjust tone or format, but can't (and shouldn't) decide your core messaging — what you stand for, why you're different, and how to position against a competitor.
- Timing and pacingAgents can trigger sequences, but pacing requires intuition.
- Creative conceptingYour campaign's big ideas — themes, storytelling arcs, and narrative tone — are still deeply human.
- Message moderationIf the task touches on strategy, voice, or brand, it needs human eyes.
If the task touches on strategy, voice, or brand, it needs human eyes.
Risks GTM Teams Flagged
The appeal of agent-led personalization is obvious: tailor every message, scale across channels, and activate the right audience at the perfect moment.
But campaigns are complex systems. And when agents move too autonomously, they can derail messaging, spam your audiences, or misfire on sensitive segments.
Put an agent on that cold call example, you could completely burn your brand's reputation if you have it start to say like nonsensical things to the prospect on the receiving end… Think about your risk-reward in the context of the job that needs to be done.
This erosion of brand trust is a huge concern. When inboxes are saturated with templated emails and AI-generated outreach, audiences are getting sharper at sensing what's real and what's automated.
Authenticity in an increasingly digital world is hard to get — and if we lean too far into AI content, AI images, AI outreach, people start to sense it. Like autotune in music, it starts to feel robotic and turns people off.
It's not just about what the agent says but also how it chooses to act. Without the right prompts or rules, it can over-personalize or push campaigns live prematurely.
Greg Baumann cautions that teams often misjudge what's truly within their control:
People are managing things at scale and they're not necessarily thinking about what's in their focus of control… If you get too dependent on the agent… it can feel like, 'Oh, the system's just doing that now,' and you lose touch with the operational nuance.
And Murali Kandasamy points to a deeper gap: Today's agents can trigger actions, but they don't know why, when, or who to prioritize.
Ideally, I want a system that can break down campaign engagement by account, visitor, and content, instantly. What's missing is something that can help us prioritize which audiences to activate, when, and with what content, based on deep engagement thresholds.
Derrick Arakaki echoes this risk of false precision, that AI can make something look scalable, when the underlying logic is brittle:
Each one (of our campaigns) feels like a snowflake as far as creating from scratch. I don't think there's a repeatable framework yet. It's just too custom to automate without putting brand risk on the line.
Big takeaway: The agent doesn't know what's high-stakes unless you tell it. The best agents in this workflow work under tight direction, pulling from pre-approved assets, scoped segments, and known triggers. But even then, they need human oversight to avoid sounding robotic, off-brand, or inauthentic. Scale is easy to automate; trust isn't.
Workflow 3
Handoffs, Follow-ups & Internal Coordination
Relevant teams: Sales, CX, RevOps
Work rarely moves in a straight line. Between every campaign, call, or customer touchpoint, there's a handoff, a baton pass between people, teams, or systems. And this is where some of the biggest leaks in GTM pipelines take place.
When it's not clear who's following up after a deal closes. When a rep forgets to add an update to the CRM. When the customer email never makes it to CX.
These coordination breakdowns — when everyone assumes someone else has it covered — are not a result of bad intent. They come from lack of visibility, repetition fatigue, and context loss. This makes internal coordination workflows one of the ripest areas for AI agent support.
What a good agent should be able to do
Agent Skill Stack
- Context capturePull action items and tasks from notes, meetings, and threads
- Follow-up nudgingRemind teammates to take action or update systems
- Escalation routingFlag risks or gaps to the right teams or stakeholders
- Cross-system syncingKeep data aligned across calendars, CRMs, and docs
- Account memorySurface prior context so teams don't duplicate or miss key info
What not to delegate to the agent
Coordination work looks structured on the surface, but a lot of it depends on judgment. Keep these human:
- Sensitive follow-ups or escalationsDon't let an agent chase sensitive customer issues or missed deadlines without human oversight. A badly timed or misworded nudge can erode trust.
- Strategic note-takingAgents won't catch nuance, intent, or emotional cues in a conversation. Humans still need to interpret and annotate what really matters.
- Cross-functional alignmentInternal team dynamics, prioritization conflicts, and high-touch customer needs still require human navigation.
Risks GTM Teams Flagged
Internal coordination might seem like the safest place to deploy AI agents. After all, they're operating behind the scenes, nudging teammates, logging tasks, syncing tools. But this is also where they're most likely to be mis-scoped. And the risks come from under-definition.
Agents operating in vague workflows with unclear ownership can create friction, add noise, and reinforce broken processes. Worse, they can create a false sense of follow-through — that something's been handled when it hasn't.
Nina Butler warns that customer trust starts to erode with inconsistent coordination, when messaging doesn't carry through from one team to the next:
Marketing must equal sales, must equal onboarding, must equal success… That story has to have continuity. If not, that's what's going to drag down your time to value and your GRRs — when you have these expectations in the game of telephone.
This is where agents should help, but only if they're scoped tightly around real dependencies and owned workflows. Greg Baumann stresses that the true value isn't in replacing human follow-through, but in nudging it at the right moment:
We've seen a lot of value in using sequences for internal prompts… People were a lot more comfortable getting that nudge: ‘Hey Greg, you owe Rahul a reply.’ That's where AI can prove its value — by prompting action, not just taking it.
But even the best nudge doesn't matter if the follow-up relies on memory. Derrick Arakaki illustrates the risk of relying on humans to fill in the gaps after the meeting ends:
You're trying to be engaged in a conversation like we are now, but I haven't written down anything as a follow-up… You'd like that to be the recap. You'd like that to be a summary you can send to the customer. But the effort to do that, you need another half hour.
Derrick envisions that's where agents — when scoped well — can step in. Not to own the customer relationship, but to ensure no part of it gets lost in the shuffle.
An agent will go out and say, hey, I have a gap here. This account, we haven't made the ask. Let me go automatically ping the CSM and ask them… Did you do this? What was the sentiment like?
Big takeaway: Agents don't fix broken coordination. They amplify whatever system they're dropped into — good or bad. If your handoffs aren't mapped, your follow-ups aren't owned, or your messaging isn't aligned, the agent won't know what to prompt, or when. But if this is done right, you get more than efficiency and continuity — and that's what drives customer trust.
Workflow 4
Post-Sale Follow-through & QA
Relevant teams: CX and Customer Success
Post-sale workflows are high-friction, high-frequency. We're talking adoption tracking, sentiment checks, QBRs, renewal prep, and putting out fires. When these processes fail, you're left with frustrated customers and missed opportunities. And no one's quite sure which accounts are actually at risk until it's too late.
What a good agent should be able to do
In post-sale workflows, AI agents are like backstage crew. They aren't speaking directly to customers; they're prepping the people who do. A well-scoped agent should flag risks, prep materials, and surface insights to help CX and Success teams stay proactive.
Agent Skill Stack
- Data parsingTrack product usage, renewal status, and ticket volume
- Summary generationCreate QBRs, risk briefs, renewal prep decks
- Sentiment detectionMonitor tone shifts in tickets, surveys, and customer calls
- Follow-up promptingRemind reps to take action on flagged issues or missed steps
- Context surfacingPull past insights to inform current escalations or planning
What not to delegate to the agent
When customer trust is on the line, certain calls must stay with a human.
- Renewal negotiations & pricingAn agent can prep the context, but the nuance of commercial conversations requires human discretion.
- Sensitive support escalationsAuto-escalating is helpful; resolving without empathy isn't.
- Strategic upsell planningAgents can flag opportunities but deciding what to offer and when is still human-led.
- QBR storytellingDrafts help. But connecting value to business goals needs human framing.
- Final call summaries & customer commsLet agents prep the pieces, but keep the trust-building touchpoints human.
Big takeaway: Let agents prep the pieces, but keep the trust-building touchpoints human.
Risks GTM Teams Flagged
AI agents in post-sale workflows can break quietly. Unlike in marketing or sales, where issues are obvious, a broken agent in CX often flies under the radar: until a missed renewal, silent churn, or unflagged risk catches the team off guard. And by then, the damage is already done.
Derrick Arakaki points out how fragile renewal workflows can be when critical cues go unnoticed:
When I think about an account that's coming up for renewal… what is our engagement with the customer, what level we're engaging with, product champions or executives? Did we make the customer aware that there's a renewal coming up? Which could be risky, right? Because I didn't know.
If an agent is supposed to track renewal readiness but fails to prompt — or worse, prompts incorrectly — that's a customer lost, not just a task missed.
Murali Kandasamy adds that even when data is available, what gets surfaced is often the wrong thing:
I want to know who's logging in, who's creating, what content they're engaging with. And more than that, what they care about. Are they looking at case studies? Are they bouncing? That's the stuff I want surfaced before a call.
That gap between raw data and real insight is exactly where agents misfire when they're not aligned to the team's judgment criteria and important “soft signals” in customer retention.
Even when agents do support prep, they often fall short when strategic nuance is required. As Derrick explains:
For enterprise QBRs, you could take two to three weeks to prepare. And you're constantly fine-tuning. AI benefits the mundane, but QBRs aren't mundane.
Agents can support, but they can't interpret politics, tone, or account history. And in high-stakes CX conversations, generic is dangerous.
Big takeaway: Agents can make post-sale workflows faster, but they can also make them blinder. Without human oversight, process clarity, and clearly defined signals, AI agents may create friction that takes months to repair.
Across all four workflows, one thing's clear: AI agents work best when they're scoped clearly, monitored thoughtfully, and matched to the right job. In the next section, we'll cover how to design for that with the right human-in-the-loop oversight, evaluation criteria, and guardrails.