{ "video": { "id": "ITMXwI6QL6A", "title": "How Forward Deployed Engineering is done at Ramp — Leo Mehr", "duration": 845, "upload_date": null, "channel": "AI Engineer", "source": "AI Engineer", "event_format": "In person" }, "analysis": { "video_id": "ITMXwI6QL6A", "title": "How Forward Deployed Engineering is done at Ramp — Leo Mehr", "one_liner": "Ramp's FDE director Leo Mehr argues forward deployed engineering rests on exactly two principles — always be scoping (interrogate the request before building it) and scale with tokens (replace each stage of the FDE pipeline with agents) — and that having only one of the two either burns weeks on the wrong feature or leaves you with a \"token maxing slop cannon.\"", "summary": "Leo Mehr, a director of engineering at Ramp, describes FDE there as living inside the engineering org with the goal of helping Ramp win upmarket, working on the core product and new agentic features for the largest enterprise customers. He rejects the idea that an FDE's job is to say yes to customers: the job is gathering enough context — what's driving the urgency, who else in the pipeline benefits, what workarounds exist — to build the right thing. He then argues that scoping mastery alone is no longer enough, and walks through Ramp's Notion-agent triage bot for their #FDE-requests Slack channel as a first working piece of an agent \"factory\" that will eventually cover the whole context → scope → spec → implement pipeline.", "key_points": [ "FDE at Ramp was two engineers 2.5 years ago; Mehr's org is now ~30 engineers across four teams — deployed, developer, API, and a new AI services business — and sits within engineering, not go-to-market.", "He explicitly rejects the meme that FDE is the \"final evolution\" or \"boss mode\" of technical go-to-market roles: that may be true elsewhere, but not at Ramp.", "The Friday-night SAP S/4HANA example: the default engineering reflex is to go find the SAP API docs; the trained FDE first asks what's driving the urgency — he's seen reps go crazy at end of quarter to hit quota, not because the customer is urgent — plus who will use it, whether workarounds are exhausted, whether a manual stopgap exists, and whether the customer can just hit Ramp's API.", "The most important scoping move is looking beyond the single request to other prospects in the pipeline and existing customers who'd benefit from the same thing.", "The mobile reimbursement failure: with the mobile team swamped, two FDEs learned iOS and Android development and grinded for a couple of weeks to ship both platforms — then the customer revealed it mandates iOS devices for all employees. Even the most basic assumptions need validating.", "Ramp's #FDE-requests Slack channel collects blockers posted by account managers, solutions and sales reps as Notion workflows, with super high variance — some are detailed customer write-ups, others are one line like \"we need this SAP integration.\"", "The V1 agent, built with Notion agents, literally just took the request and asked a couple of questions; within a couple of weeks reply latency dropped from hours or days to seconds and reps started engaging with it. The current iteration (with a penguin avatar to feel friendlier) runs several rounds of back-and-forth with the submitter until it judges the request ready for a spec — saving roughly 20% of scoping time.", "Pipeline status: the first stage (context gathering/scoping) now has an agent; the last stage is easier because frontier models can one-shot medium-size features; the middle — spec shaping — is \"gnarly, unformed and difficult\" and is where the team plans to invest.", "The failure modes cut both ways: a factory without scoping discipline produces a \"token maxing slop cannon,\" while great scoping without an agent factory means agent-native competitors outcompete you." ], "takeaways": [ "Before writing a line of integration code for an urgent enterprise ask, establish what's actually driving the urgency (quota vs. customer), whether workarounds or a manual stopgap exist, whether the customer can use your API instead, and which other pipeline prospects would benefit.", "Validate the most basic assumptions of a request — even which mobile platform the customer's employees are allowed to use — before committing engineering weeks.", "Map your own role as a pipeline (gather context → scope → spec → implement) and attack it stage by stage with agents; a trivially simple V1 that just asks the submitter a couple of clarifying questions can pay off within weeks by collapsing reply latency from days to seconds.", "Invest in the applied-AI infrastructure around the agents: a smooth agent harness, output quality measured with evals, rubrics and human feedback, and getting the right context in — the tacit product knowledge in a PM's head that Notion docs and help articles only partly capture.", "Do both principles together: keep human taste and judgment over the final output, because a token factory fed by unscoped requests just produces slop faster." ], "topics": [ "forward-deployed-engineering", "agents", "scoping", "enterprise", "evals", "context-engineering", "workflow-automation" ], "tools": [ "Ramp", "Notion", "Notion agents", "Slack", "SAP S/4HANA", "Waymo", "iOS", "Android" ], "quotes": [ { "text": "So, unless you are scaling with model capabilities, you are going to fall behind.", "at": "06:33", "url": "https://www.youtube.com/watch?v=ITMXwI6QL6A&t=393s" }, { "text": "they mandate all of their employees to use iOS devices.", "at": "05:49", "url": "https://www.youtube.com/watch?v=ITMXwI6QL6A&t=349s" }, { "text": "you're going to get a token maxing slop cannon.", "at": "12:53", "url": "https://www.youtube.com/watch?v=ITMXwI6QL6A&t=773s" }, { "text": "as an FD, we still have the responsibility of taste and judgment over the final output.", "at": "12:22", "url": "https://www.youtube.com/watch?v=ITMXwI6QL6A&t=742s" } ], "words": 2760 }, "summary_url": "/#ITMXwI6QL6A", "transcript": { "html": "/transcripts/ITMXwI6QL6A.html", "txt": "/transcripts/ITMXwI6QL6A.txt", "vtt": "/transcripts/ITMXwI6QL6A.vtt" } }