Welcome to edition #39. I sat down with three launches this week and tried to write them up as three stories.
They are one story.
Clay moved into orchestration. Seamless moved into enrichment-by-prompt. Clari moved into email marketing. Same week, same move.
Then on Monday, Seismic and Highspot finished merging into one company.
So it is happening in two directions at once. Layers eating the layer next to them, and peers collapsing into each other. Either way the count goes down. Two things to sit with. One, if your glue lives in n8n or Zapier, the company that owns your data is now selling glue. Two, the feature you chose a vendor for last year is turning into a checkbox on somebody bigger’s roadmap, or into a line item in somebody’s merger FAQ.
Let’s dive in.
One date before we start: Cohort 5 enrollment closes 14 September. The five roles further down are the seats it trains for.
Recent news
Clay ships Workflows, and the company that owns your data starts selling the glue too. Clay launched Workflows on 11 August: a node-based canvas where triggers, conditional logic and agents run against your Audiences data layer, with every run traceable to the data points it read. Build one from the CLI or describe it to Sculptor. One closed-beta customer has already moved a workflow off n8n.
Seamless.AI ships AI Smart Fields, and the AI column stops being a differentiator. Seamless launched AI Smart Fields on 13 August: write a prompt, get a generated column across contacts, companies, or a whole list. That is Claygent’s primitive arriving in an incumbent sales-intelligence platform. Five days later it shipped an MCP server for Claude too: stop being a place people visit, start being something an agent can call.
Clari + Salesloft absorbs AI email marketing, and an agent qualifies pipeline before a human sees it. Clari + Salesloft announced a partnership with allGood, whose agent Mary reads inbound replies, determines intent and routes prospects without a rule tree behind it, already in production at Zendesk, Bonterra and Netskope. Customers on the legacy email product are pointed at a migration guide: a replacement, not an add-on.
Seismic and Highspot complete their merger, and the category admits there was room for one winner. The two largest sales enablement platforms closed on 18 August and now operate as one company under the Seismic name, led by Rob Tarkoff: 2,500 customers, 3.5 million users. Unlike the three above, this is two peers becoming one rather than a vendor absorbing the layer next door.
Favorite MCP of the week
6sense MCP Server. Announced 10 August and now generally available, it makes 6sense intelligence callable from any MCP-compatible agent: Claude, ChatGPT, Writer, Agentforce. Account insights, predictive buying stages, 6QA status, keyword intent and ad campaign performance, with no custom integration.
Look at what that actually is. 6sense’s moat was proprietary intent data, held inside a platform you had to log into. The MCP server takes that data out of the dashboard and puts it in the prompt. It is the same move as the three news items above, run on the intelligence layer instead of the workflow layer: stop making people visit your interface, start making your data reachable from wherever the work happens.
One catch, and it is a real one. It needs a 6sense subscription, so this is not one you wire up on a free tier this afternoon. But it is a good signal of where the incumbents are heading, and a reasonable thing to ask your other vendors for.
LinkedIn hot takes
“The agent is allowed to be wrong. It is not allowed to send.”
Mani Iyer runs an agentic GTM platform and used his own category’s worst week to make a point nobody selling AI SDRs wanted made. His numbers: teams pushed 6.4x more outbound, from roughly 1,150 to 7,360 touches per rep per month, and positive replies fell from 2.1% to 1.3%. Half of all pilots shut down inside 90 days. One tracked deployment hit 40% deliverability by week three and took eight weeks to recover. His read is that this was a design failure rather than a technology one: the agents started from lists instead of signals, deleted human judgment, and burned the sending domains. Read it in full if somebody is selling you an autonomous SDR this quarter.
“Building the automation took an afternoon. Writing down how the business actually decides took 3 weeks.”
Christina Koch handed Claude Code her whole GTM motion and it runs end to end, but the post is not about the automation. It is about the three files underneath it: icp.md with the profile and buying signals, scoring.md with deterministic tier rules, and copy.md with the frameworks that have actually booked meetings. Claude does not improvise. It reads the files and works inside them, and results write back so the next run starts smarter. Her closing line is the one to steal: the context layer is the product, the automation just executes it. If you cannot write your ICP and scoring rules in plain English, no model fixes that for you.
“Adoption is the only metric. Everything else is a demo.”
Anis Nanai, founding GTM engineer at Owl Bound, on why GTM systems fail. Not in production, and not because of tool churn. They fail because the AE does not trust the score, opens Sales Navigator, and does it their way. The build works, the motion does not change. His estimate is that the build is 30% of the job and getting people to understand it, trust it, and keep using it after you stop watching is the other 70%. Every project is also a negotiation between four definitions of working: sales wants meetings this week, marketing wants attribution, RevOps wants clean data, leadership wants one number.
“Nobody wants to click around a dashboard.”
Adam Schoenfeld tracks 743 GTM brands and logged 23 major moves last month, which makes him the closest thing this category has to a scoreboard. Five acquisitions in July alone: Common Room to Zoom, Ciro to Reevo, Seam AI to Clarify, Commsor to The Swarm, Scout AI to Orum. Two more in the first week of August. And the line that matters for this issue: Clay, Apollo, Webflow, 6sense, ZoomInfo, Clari, HubSpot and Outreach all shipped something an agent can call. CLIs for GTM grew 50% month over month. If everything above in this issue reads like a pattern, that is because it is one, and somebody is counting.
GTME jobs
★ Star pick. GTM Engineer @ Mercor | $175,000 to $250,000 | San Francisco, onsite Building the go-to-market engine from scratch: data pipelines, integrations and AI agents across sales, marketing and customer success. The stack is the tell. Clay, Attio, Snowflake, BigQuery, dbt, Python, SQL. That is a data engineering job with a GTM surface, priced accordingly, and it is what the role looks like when a company decides it is an engineering discipline rather than an ops one.
GTM Engineer (Data & AI) @ LeanData | $120,000 to $170,000 | United States, remote Unified data layers and agentic workflows that reason through signals to automate GTM decisions. MCP is named in the stack, which is still rare in a job description. LeanData built its business on lead routing, so this is the incumbent rewiring itself around agents rather than a startup arriving with them.
GTM Engineer @ Pogo | comp not listed | United States and Canada, remote First GTM engineer at the company, and the widest stack on this list: Clay, HubSpot, Apollo, Outreach, Salesforce, Claude, Codex. LLM agents for lead routing, enrichment, outbound personalization, account research and attribution. Being first means you set the architecture rather than inherit it.
GTM Engineer @ Fireworks AI | $140,000 to $170,000 | San Francisco Bay Area Building the GTM tech stack behind the sales engine at an AI inference company. Salesforce, HubSpot, Gong, Clay, Zapier. Integrations, automation and workflows that shorten the path from signal to seller.
Forward Deployed GTM Architect @ Artisan | comp not listed | San Francisco, onsite The most interesting title of the week. Architecting GTM systems for Artisan’s largest enterprise customers, spanning CRM architecture, RevOps alignment, compliance and hands-on deployment. Clay, Outreach, Salesforce, SQL. The forward-deployed engineer pattern has arrived in GTM, which is its own signal about how much customization this work still takes.
Cohort 5 of GTM Engineer School is open, and the five roles above are the seats it trains for. Four of the five name Clay. The fifth names MCP. One of them asks for Claude and Codex by name. That is the job now, and it is the job the program builds you into. Eight weeks at six to ten hours a week: four hands-on in Clay, Octave, Swan and Sumble, taught by the teams who built them, with Claude running through all of it and thousands of dollars of platform access and credits included. Then four weeks building a capstone for a real customer on your own stack, presented at demo day with recruiters in the room. 265 operators have come through the first four cohorts. $1,850. It starts Monday 28 September, and enrollment closes 14 September.
Top GTMEs to follow
Joe Rhew runs gtmengineersearch.com and publishes The Shortlist, a recurring post of net-new GTM engineering roles with real comp bands and stacks. He also published the role taxonomy a lot of us now use to describe the work.
Elias Stråvik open-sourced
/gtmskills, plus a multiplayer context setup and a Slack-deployable GTM agent. If you want to see what portable GTM logic looks like in practice rather than in a diagram, start there.Benjamin Reed, founder at RevyOps, works on signal and intent plumbing. The unglamorous layer that decides whether everything above it works.
Brigitta Ruha, co-founder at Growth Today, on allbound motion and AI-native GTM for teams that do not have a twenty-person revenue org.
Nikhil, fractional GTM engineer building AI-native systems for devtools and AI SaaS, previously at Pieces. Writes about the parts nobody demos: identity resolution, retry logic, and what happens when the data is wrong.
Recommended resources
“The State of LinkedIn Outreach H2 2026” (Expandi). 13.2 million connection requests and 6.7 million messages, and one finding worth the whole report: compared within the same account, AI-hyperpersonalized messages were accepted 12% less often than that account’s own human-written templates. Reply rate was flat. Cross-account numbers say the opposite, and the gap is selection bias, because stronger operators adopt AI first. Also useful: three-message sequences reply at 9.8%, and five or more drop below sending a single message.
“gtm-starter-kit” (The Revenue Architects, MIT). The repo version of what Christina Koch is describing above. A
CLAUDE.mdplus context files for ICP, signal library, positioning and personas, and skills that read them rather than improvise. The part to actually open isexamples/sample-company/, which is fully populated, so you can see what a mature signal library looks like, decay logic and all, before writing your own.“GTM Engineer Salary 2026: Real Ranges From 491 Hand-Checked Roles” (Steven Moody). First-party research rather than a survey write-up: 491 US reqs checked by hand with agencies and duplicate postings stripped out, plus 28 hiring-team interviews. An $84K gap sits inside the single job title, and the finding worth arguing with is that roles naming Clay pay around $17K less, because naming the tool marks the job as outbound rather than engineering.
One last thing. Everything above says the same thing: the vendors are collapsing into each other, and the durable position is being the person who wires them together rather than the person who bought them. That is the whole design of Cohort 5, and it is why the five roles further up read the way they do.
You spend the first half hands-on in Clay, Octave, Swan and Sumble, taught by their own teams rather than by someone reading their docs, and in Claude, which is where you wire them into something that actually runs. You spend the second half building a system for a real customer on the stack you already run. The tools are the vocabulary. The system you can design with them is what transfers, and that is the part you keep.
Enrollment closes 14 September.
Which of your tools is about to become somebody else’s feature? Reply and tell me, I read them all.
If this was useful, forward it to the person on your team who owns the stack.
Jared