Month Two: The FDE Leaves, the Workflow Has to Stay

Gartner put a number on month two last week. On September 29 it predicted that by 2028, 70% of enterprises will abandon agentic AI built by vendor forward deployed engineering, "trapped by soaring costs and unable to evolve it on their own." It's a forecast, not a measurement, and I'd hold the 70 loosely. The second half I take seriously. Unable to evolve it on their own. That's a handoff nobody made.

Mukul Saha, the Gartner analyst on the release, says FDE success "starts with getting the engagement structure right, from scope and incentives to governance, ownership, and exit." The release tells buyers to run the exit plan they set at the start "rather than extending the engagement because internal teams are not ready." Matt Ashare at Channel Dive reported the price of not being ready: Gartner estimates FDE fees alone "could top $200,000 quarterly per use case."

That's the buyer's side. Two pieces from September describe the same hole from the vendor's side.

Simon Frey, Gong's chief customer officer, wrote on September 15 that AI is exposing post-sales more than fixing it. He lays out the chain: "Sales closes the deal, onboarding gets the customer started, a CSM manages the relationship, renewals handles the commercial motion, and support steps in when something goes wrong." Then the line I keep rereading: "Every handoff created an opportunity to lose context." And on the seat this series is about: "Without a clear operating model, FDEs risk turning an organizational problem into an engineering project."

Eleven days earlier Raj Bheda, co-founder of GainTrace, named the exact spot. His split: "the FDE owns the technical outcome at deployment, the CSM owns the business outcome across the term, and the handoff between them is the point most companies never write down."

Both are vendor blogs. Both are right about this.

This is part 6 of The FDE Transition. Part 5 was the contract with sales, before the signature. This part is the other end: month two, when the FDE leaves and customer success takes the account.

The short version

The FDE's attention is what holds a new deployment together, and at day 60 that attention drops from one of three customers to one of twelve accounts, held by someone who can't read the configuration. So the handoff has to be a page, not a meeting. Four lines: the promise, which is the handshake's three columns updated; the signal, which is the customer's eval rows on a schedule with the go-live pass rate as the floor; the watch list of what this customer is likely to switch off; and the keys, the named person at the customer who can change the customer layer without calling you. Plus a recall rule that says what brings the FDE back and how fast. The CSM inherits the page and owns the account's trajectory. The CSM does not inherit the keyboard. The overlap starts at go-live, and if the keys line is blank at day 60, the deployment hasn't landed.

Why month two is where it breaks

Jason Lemkin's SaaStr answer from March has the arithmetic. A CSM, in his description, runs a portfolio of 8 to 12 accounts. An FDE embeds with 1 to 3 customers, and a single deployment can take 30 to 60 days or more. "They're barely the same profession."

For two months the customer was a third of somebody's working week, and that somebody could open the config and fix the thing by lunch. On day 61 the customer is a twelfth of somebody's week, and that somebody can see a health score.

Part 3 has a day-60 reversal update, the note to product on what the customer undid once the FDE stopped watching. I wrote there that it tells you whether the delta was real or whether the FDE's presence was holding the workflow together. That's the month-two question from product's side. From the customer's side it's simpler. Does the thing still work when nobody who built it is in the room?

I've lived the general version. In a pricing migration I ran, 12 accounts sat on plan on the dashboard while their usage ran under the committed minimum. Nobody escalated. We lost two of them. The behavior moved first and the dashboard never did.

Bheda's table shows why the gap exists. The FDE is measured on "time to live, technical milestones shipped, whether the deployment works." The CSM is measured on "gross and net revenue retention, renewals on time, expansion sourced, and the health of the book." Neither list has the thing in between on it: is the customer still running the workflow the way it went live.

The month-two page

One page, written by the FDE, signed by the FDE and the CSM at day 60. Four lines.

The promise. The three columns from the part 5 handshake, updated. What shipped. What's still in the builds column, and the date. And the no's, because the customer's champion will ask the CSM about one of them in month four and the CSM needs to know it was a no before the signature, not a slip after it.

The signal. The customer's eval rows, the ones from the pre-contract week and the handback packet, running on a schedule against production. The pass rate at go-live is the floor. Frey's piece says email response rates for NPS and CSAT surveys fell from 20 to 25 percent in 2019 to 10 to 15 percent in 2025 (his cited figure, not mine), and his advice is to watch real customer signals instead. For an agent deployment the real signal already exists. It's the rows. A CSM who gets a weekly pass rate on the customer's own cases knows more than any survey will tell them, and knows it sooner.

The watch list. Three or four things this customer is most likely to switch off, override, or route around, and what each looks like in the observation record. The FDE knows these. The approver who never liked the auto-routing. The team that kept the spreadsheet open next to the agent for the first three weeks. Written down, a reversal becomes something the CSM can see coming. Unwritten, it's a surprise at the renewal.

The keys. The name of the person at the customer who can change a prompt, a rule, or a threshold in the customer layer without calling you. This is the line that matters most, and it gets the next section.

Then one rule under the four lines: the recall. What brings the FDE back, and how fast the FDE's manager has to answer. I'd make it three triggers. The rows fall under the floor. Something on the watch list happens. Or the customer wants a new workflow, which is the expansion part 5 said belongs to the FDE. And I'd put two business days on the answer.

Who holds the keys

Part 4 gave the FDE the customer layer outright: prompts, rules, thresholds, connector settings. Same-day changes, no review. So who gets that layer when the FDE walks out?

Not the CSM. Chad Horenfeldt, who managed a team of FDEs back in 2016 and now works at an AI-native company, called FDEs-as-CSMs a trap in August 2025. His question: CSMs should be technical enough to advise, "but should they be the ones actually creating the webhooks and troubleshooting poorly created AI prompts?" And his answer: "Getting into the weeds of solutions takes CSMs away from taking the time to understand their customers' outcomes." I agree. A CSM with twelve accounts and edit rights on twelve configurations is a worse FDE and a worse CSM.

The keys go to the customer. One named person on their side who can make a layer-one change and see the rows run afterward. That's my test for whether a deployment landed. If the only person who can change it has left the building, the customer has a dependency, and the FDE was holding the thing up. That's Gartner's sentence at the size of one account. "Unable to evolve it on their own" is a blank keys line, two years on.

The handbook's org design chapter says adoption is a deliverable, owned by name, and that it isn't the same deliverable as the build. The keys line is how you check that it was delivered. It's also the honest version of the Decagon argument from part 4. The question was never whether you need an FDE. It's whether the customer still needs yours in month three.

Harness changes don't move. They still go in as an FDE pull request with a price, whoever asks.

Overlap at go-live, not at day 60

Frey's fix for the chain is one owner: "Give one person accountability for the account's trajectory, then use technical specialists and AI agents to support that ownership at scale." I'd take that with one amendment for this seat. Before day 60 the owner is the FDE. After it, the CSM. And the two overlap for the whole stretch in between.

So the CSM joins the week the customer goes live. They sit in on the FDE's check-ins, watch the workflow run, and meet the person who's going to hold the keys. By day 60 the page is a record of things the CSM has already seen. Start the handoff the day the FDE leaves and the CSM gets a document. The context stays in the FDE's head, and the head is at the next site.

And every recall is data. One customer recalling is support. Three customers recalling for the same reason is the count from the handback packet, and it goes to the product owner who reads those. The economics chapter says FDE hours per deployment should fall every quarter. Recalls per deployment should fall with them. Gartner's second prediction in the same release is that through 2028 less than 20% of FDE engagements will turn recurring customer needs into capabilities in the vendor's core product. Recalls are where recurring needs show up after the FDE has gone. Count them or you're in the 80.

What this means in the seat, and at the top

If you're the FDE: write the page in week six, not on your last day. Pick the keys person early and make your last three config changes with them driving.

If you run customer success: ask for a seat at go-live and a weekly pass rate on the customer's rows. Turn down the edit rights.

If you run the company: no deployment counts as landed until the keys line has a name on it that isn't yours.

One thing to try this week: pick a deployment where the FDE rolled off more than 60 days ago. Ask the CSM four questions. What did we promise, column by column. Which of the customer's eval rows still run. What has the customer turned off. Who over there can change a prompt. Count the blanks. That's the size of the handoff nobody wrote down.

Sources: Mukul Saha, "Gartner Predicts 70% of Enterprises Will Abandon Agentic AI Built by Vendor Forward-Deployed Engineering by 2028," Gartner (September 29, 2026) · Matt Ashare, "Are forward deployed engineers flirting with failure?", Channel Dive (September 30, 2026) · Simon Frey, "AI won't fix post-sales until you fix the operating model," Gong (September 15, 2026) · Raj Bheda, "Will Forward Deployed Engineers Replace CSMs? What Each Owns," GainTrace (September 4, 2026) · Jason Lemkin, "Dear SaaStr: Can I Convert My Customer Success Managers into Forward Deployed Engineers?", SaaStr (March 26, 2026) · Chad Horenfeldt, "Forward deployed engineers as CSMs: It's a trap," Customer Success & Failures (August 26, 2025)

Part of the running argument on AI Product Management: once building is cheap, the scarce work is deciding what gets built for everyone, and month two is where you find out whether what got built for one customer stays built.

Related answer: Who owns the customer after a forward deployed engineer leaves?