Skip to content
LAST REVIEWED 2026-11SOURCED FROM 4 SESSIONS, MAR 2026 ONWARD

The ground truth: GTM is the hard part now

Section titled “The ground truth: GTM is the hard part now”

A multi-exit founder’s session on the shifting economics of software framed the decade: product and R&D costs are collapsing (idea to production in days, not quarters) while distribution gets harder and more expensive. Seat-based pricing is eroding wherever the end user is an agent rather than a human; buyers research in AI assistants before ever visiting your site.

The practical to-dos that fell out of that session: optimize to be recommended by AI assistants, ahead of ranking in search; ship API-first and expose your product where developers’ AI tools can find it; and use agents for the funnel’s expensive edges: lead generation that cost ~$100 a lead through agencies runs near $1 with an agent stack, though closing five-figure deals still takes humans.

The strategy those tools are supposed to serve (five questions, a lane you own, a funnel that is now a scatter plot) is GTM as a system.

One salesperson with an AI skill is actually a lot more valuable than maybe a 40 BDR team.

Multi-exit SaaS foundersession, Apr 2026

The agency team’s core buying rule: choose marketing software by API robustness, because the future is wiring your tools together with AI coding assistants into small purpose-built apps, not accumulating subscriptions.

It’s a fantasy that we’re gonna replace entire marketing teams with this stuff anytime in the near future.

Marketing agency co-founder, ~20 years in marketingsession, Mar 2026

Their working patterns:

  • Synthetic-customer testing: a micro-app that generates a hundred test customers matching your segment, crawls your site, and reports what would stop them from buying: a cheap first pass before real customer research.
  • Fast message validation on LinkedIn: post variants, watch what resonates, double down. One founder effectively built the product from the reactions.
  • Big A/B surface: hundreds of ad variants with agents auto-killing losers.
  • Context over prompts: “prompts are easy… it’s all about context.” Day one of an engagement is collecting API keys so the AI can see real CRM and analytics data.

An October session with a repeat founder and AI go-to-market operator added two useful ways to organize the work. One is to give an existing teammate a reusable skill for a task such as proposal drafting or renewal preparation. The other is to assign an AI role a continuing mandate, a defined scope, access to specific tools, and feedback it can retain. The speaker uses both; these are working models, not a standard technical definition of an agent.

Start by mapping the workflow as it runs today: inputs, decisions, handoffs, time spent, and the output that matters. Pick a bottleneck with enough business value to justify improving it. Mark which steps need human judgment or relationships before deciding what to automate.

Shared context matters as much as the role. Keep account history, source documents, commitments, and reusable instructions organized so the next task does not start from an isolated chat. Capture meeting context when appropriate and permitted; when recording is unsuitable, write useful notes promptly. A proposal should be grounded in what the customer actually said and what the team can deliver.

Three ways to organize AI work in a go-to-market team

Section titled “Three ways to organize AI work in a go-to-market team”

In a follow-up session, the same operator kept the two October models and added a third. He said the right one depends on the team: a Series A company hiring sales development reps and account executives has a different problem from a founder doing sales alone who wants to operate like a larger team.

  • Supercharge the people you have. Give each person skills, which he described as saved recipes or programs for work they already do well, so one rep can have the impact of several.
  • Add AI teammates. Each teammate is a role on the org chart with its own chat account, a job description and only the tools that job needs. When one makes a mistake, he sends it private feedback and it learns. His event-planning teammate has no access to production code, just as a human event planner would not.
  • Give everyone an AI chief of staff. A generalist assistant for each team member that watches calendars and transcripts so nothing gets dropped.

It’s a mental model very much like managing a person.

Repeat founder and AI go-to-market operatorfollow-up session

He uses the first two together and saw no reason not to use more than one. Before building any of them, his team starts client work by auditing workflows: where the time goes, which tasks recur daily or weekly, and which have the most effect on the business. With one go-to-market team of about 50 people, a survey of reps about where their time went pointed to a data quality problem in the CRM. His recommendation was to stop other work and build one team-wide agent that, at the end of each day, moves the call transcripts into the CRM, which he expected to improve campaigns, sales performance and rep briefings at once. The second workflow they found was RFP response.

The operator’s view was that agents keep getting better, but they cannot make up for missing or scattered data. His checklist for a go-to-market team:

  • Keep full transcripts, not only AI summaries. “AI notes, in my opinion, worthless,” he said. “What you want and need is the transcript.” A transcript shows who said what and lets you replay the clip. He said the best sales teams have both real-time transcription they can work with during the call and the recorded artifact afterward.
  • Connect the notetaker at the workspace level, so sales leaders can look across every call on their team, not only their own.
  • Use one notetaker across the team. If some reps use one tool and others use another, the split shows up when you try to search across all calls.
  • Audit connectors. His first check with a go-to-market leader is which core systems are connected to their AI tool, with what read and write access. Missing connections mean the assistant works from an incomplete picture.
  • Let AI keep the CRM current. A Series A founder in the session had an agent fill in the qualification framework after each call from the transcript, because junior reps often forgot. The operator said that if you are not using AI to update your CRM, “you are behind.”
  • Add an acceptable use policy once you hire at scale. He said small teams rarely need one.

He also said where this shared context lives is unsettled. His team had tried a code repository and found it was not built for the job, moved to a docs workspace plus shared drive, and was about to evaluate a data lake, because the most advanced companies he had come across pulled everything into one.

Define access, approval, and cost for each workflow

Section titled “Define access, approval, and cost for each workflow”

The October operator described giving different AI roles different access. A content or event role does not automatically need accounting or CRM permissions. Define the systems and actions required for the task, along with the person responsible for the result.

The session also described an outbound assistant sending a message despite instructions to wait for approval. In the follow-up session, the operator said part of the cause was his own choice: to save cost on what seemed a simple drafting task, he had switched that role to a smaller model, and the judgment was not there. The lesson is that a written instruction alone did not prevent the action in that setup. The approach the operators described is to separate drafting from sending or publishing, and to enforce the approval step in the tool or workflow before granting external action privileges. In their experience, a role that can prepare material is not automatically ready to contact customers.

Measure total workflow cost, including usage charges, subscriptions, review time, and rework. The speaker tracked spending by person, role, and workflow after agent usage added costs beyond existing subscriptions. Compare that cost with useful work completed; changing a model or reducing how often a task runs also needs a check on output quality.

This is an editorial template based on the session, not a promise of autonomous performance.

AI GTM WORKFLOW BRIEF
Workflow and business outcome:
Current steps, owner, and time required:
Inputs and source of truth:
Human decisions or relationship work to preserve:
AI task or continuing role:
Systems it needs to read:
Actions it may take without approval:
Actions requiring enforced human approval:
Reviewer and quality checks:
Usage budget, review time, and rework to track:
Success measure and review date:

See Building a repeatable sales motion for using this approach in qualification, handoffs and account preparation, and Marketing an AI product for preserving the founder’s voice.

Start AI teammates on narrow, scheduled jobs

Section titled “Start AI teammates on narrow, scheduled jobs”

Asked how he manages time across several AI teammates, the operator said he focuses them on scheduled work he trusts and uses skills for one-off tasks:

  • A newsletter that goes out every week, drafted every Tuesday.
  • A morning sweep of things he should not drop. His chief of staff role had sent a sweep three times a day, which got too expensive, so he cut it to once each morning.
  • Five topics to write about, every Friday.

He said there is no need to answer bots immediately. Agents will also message each other and act in shared spaces. When his sales teammate disagreed with how he handled a prospect, it posted its objections in a public revenue channel. His fix was the feedback he would give a person: he told it to give criticism privately, not in a public channel.

Two further examples from the session:

  • Consolidating bots. A sales leader with nine direct reports built a narrow bot for each report to capture one-on-ones and chat exchanges. Rather than chat with nine bots all day, the operator’s team set up one chief-of-staff agent that talks to all nine and sends the leader a single summary before each meeting.
  • Support with escalation. The operator’s own support agent, trained on years of past tickets, answers basic requests such as password resets and escalates refunds and angry customers to a person. Another operator he described handles about 200 support tickets a day with three domain agents and a manager agent. About 90 percent of tickets are fully automated, the manager handles cases such as refunds, and the hardest cases go to a person.

A small team with few agents set up planned a targeted campaign to about 100 hand-picked prospects and asked whether one agent could run all of it. The operator suggested first having some evidence that the strategy is the one you will stick with. Then split the work: one agent builds the personalized demo instances, and a separate agent handles outbound email, which should come from the founder or a named account. Start with the agent drafting the sequence, run it on lower-stakes prospects as confidence grows, and only then automate more. He also said the team would want AI access to its own product, through MCP, in every scenario, and should move it up the roadmap. For mission-critical work he uses the strongest models. He suggested running the job side by side on a few setups, including one where you can pick the model and a plain skill, since with only 100 prospects “you only get one shot.” The founder’s takeaway was to use agents for parts of the campaign instead of doing all 100 by hand. Another founder’s was to keep it simple and avoid over-engineering.

Narrow, narrow, narrow.

Repeat founder and AI go-to-market operatorfollow-up session

A Series A founder asked how to build enablement when reps cannot manage many agents at once, and how to know the agents produce good outcomes. The operator called this “the new rev ops.” At a previous company with a large sales team, he said, performance improved once they hired an experienced revenue operations lead. He recommended that teams with the resources give one person the job of managing skills and agents:

  • Who can add a skill to the team library, and the process for doing it.
  • How skills stay current as systems are added.
  • Who prunes skills nobody uses.
  • How token spending maps to outcomes, across every AI tool, in one place.

What can your AI see? What can it do? And can it improve the system itself?

Repeat founder and AI go-to-market operatorfollow-up session

Around that owner, he listed strategy, enablement and policy: a statement of the team you want to be and what proficiency looks like by when, training, and the access rules above. To move beyond a few highly productive individuals, he said, at a minimum keep team skills in a shared system that any AI tool can point to, because the leading models keep leapfrogging each other. The founder said the company’s head of go-to-market could spend about 30 percent of their time on this, and the company was considering hiring a revenue operations person or go-to-market engineer to maintain the data and skills.

Why automated LinkedIn outreach is still risky

Section titled “Why automated LinkedIn outreach is still risky”

The operator said he had not found a good automated answer for LinkedIn outbound, because losing the account would mean losing a crucial channel. For a couple of years, an offshore agency used his account to send connection requests to his target buyers, followed by a short note asking for a first call. It cost $2,000 a month and, at its peak, produced three to five new meetings a day. Results fell as people were flooded with AI outreach, so he stopped paying for it.

He then built a guardrailed AI skill to send invites overnight, limited to 25 a day, which he called a safe number. It went over the limit, his account was flagged, and he shut it down. The AI teammate platform he uses will not do LinkedIn outbound except through semi-approved automation tools.

  • Don’t build a custom CRM. “Once you build the beast, you have to feed the beast.” Most teams use ~20% of their CRM already; archive dead contacts to stop billing creep and hire a cheap consultant instead of rebuilding.
  • Don’t put a chatbot on inbound sales. A motivated human on round-robin responds faster and doesn’t lose prospects. Treat customer-facing AI “with the sophistication of a toddler who can speak very well”: heavy context, heavy guardrails.
  • Don’t frame AI as headcount cuts. Frame it as making the existing team several times more effective: that’s also the version that’s actually true.

Both sessions converged here: instrument everything, build the live dashboard on your own product’s usage data, and treat that data as the asset: see Data moats and flywheels.

Four sessions: an AMA with the co-founder and head of growth of an AI-forward marketing agency (Mar 2026), and a talk by a multi-exit SaaS founder on how AI is repricing software and go-to-market (Apr 2026), plus an October 2026 session with a repeat founder and AI go-to-market operator on workflow mapping, scoped AI roles, shared context, approval boundaries, and workflow costs. A follow-up session with the same operator, supplied as an undated transcript on November 3, 2026, added the three working models, transcript and connector practices, scheduled AI teammates, skill library ownership and the risks of automated LinkedIn outreach. Participant and company names, and the vendor tools discussed, are omitted.