How AI Forward-Deployed Deployment Startups Actually Work

Every metric in this report is labeled by confidence: CONFIRMED disclosed by the company/primary filing · REPORTED credible third-party press · INFERRED Galileo estimate, reasoned not measured. Neither Distyl nor Northslope publishes a rate card; treat all pricing/margin figures as low-confidence unless flagged otherwise.

1.The model in one screen

An AI forward-deployed (FDE) deployment startup is a venture-backed company that embeds its own engineers inside a large enterprise, harvests the customer's proprietary context, stands up custom agentic/LLM systems against that context, and bills partly on the business outcome rather than on hours. It is the Palantir go-to-market re-run on the LLM/agent layer.

The core loop is four moves, run on repeat per customer:

1. Embed — a small forward-deployed engineering pod sits inside the customer, not in a vendor office.
2. Harvest context — ingest the customer's SOPs, systems of record, documents, and tribal knowledge; the proprietary context, not the model, is the raw material.⁴
3. Build agentic logic — convert that context into production decision systems and agents, wrapped in a verification/eval layer so the output is reliable enough to put in front of regulators and customers.³
4. Bill on outcomes — tie part of the fee to a measured business objective (cost saved, resolution time, containment), explicitly positioned against time-and-materials consulting.⁵

Why now

Three forces converged in 2024–2026. (a) Foundation models commoditized. Claude, GPT and Gemini converge on similar baseline capability, so value migrates from the model to the integration-and-application layer.¹⁴ (b) The pilot-to-production gap became the bottleneck. Enterprises had endless AI pilots and almost no production wins; Distyl's CEO frames the company as the call you make when "you're not getting enough out of your endless AI pilots."³ (c) The Palantir diaspora became hireable. A cohort of ex-Palantir deployment people now knows exactly how to embed-and-build, and the FDE label exploded — forward-deployed-engineer job postings grew 1,165% year-over-year Jan–Oct 2025, peaking in October 2025.⁹ REPORTED

2.The engagement lifecycle, step by step

This is the heart of the business. A founder is not buying a model; they are operating a repeatable embed-and-build motion. Here is how one engagement actually runs, reconstructed from Distyl's case studies, the founder's own descriptions, FDE job postings, and the Palantir lineage the model copies.^1,3,4,9

Phase 1 — Discovery & partnering "from day zero"

Unlike a SaaS sale, the entry is consultative and relationship-led. Prakash's framing: "Distyl partners with leaders from day zero to design that transformation, embed engineering talent, and deliver outcomes within three months."¹ The deal is co-designed with a senior executive sponsor, and the FDE team works to identify the highest-value, most checkable workflow to wedge into — not a flashy demo, but a high-volume, currently-expensive, measurable process.

Phase 2 — Scoping the wedge workflow

The pattern across all six public Distyl case studies is the same: pick one concrete, repetitive, decision-heavy workflow with a defined "done" state. Examples: prior-authorization decisioning, supply-chain root-cause analysis, loan-origination fraud review, provider-contract extraction.¹² Crucially, success metrics are defined up front (cost saved, accuracy %, resolution time) because the contract will be priced against them.

Phase 3 — FDE embed: what they do day-to-day

The FDE is the unit of production. Per a 1,000-posting analysis, the dominant "Builder FDE" archetype (≈60% of all FDE roles) is 70–90% coding, 30–50% travel, mid-to-senior, paid like an engineer not a salesperson.⁹ Inside the customer, a Distyl FDE's week looks like:

• Working directly with customers — the #1 responsibility in 55% of FDE postings; sitting with SMEs to learn the actual workflow and its edge cases.⁹
• Building & deploying AI/ML systems (37%) and integrating systems/APIs (32%) — wiring the agent into the customer's data via SQL, APIs and document repositories.^9,4
• Owning the live system — monitoring, debugging, observability, and iterating against real-world feedback; "own an entire deployment when things go sideways at 2 AM."^7,9
• Feeding the platform team — representing the customer's voice back to the central Distillery product org so field-learned patterns get productized.⁷

Phase 4 — Build, validation & handoff

The build is not "done" until it survives validation against the customer's own ground truth. Distyl's selling point to SMEs: the platform writes down its step-by-step reasoning and source citations, not just the answer, so clinical/finance experts can audit and refine the logic in natural language.^11,12 Prakash is blunt that this human-in-the-loop step is non-negotiable: "Without the subject matter experts, there's no chance we're able to go into production."³ Handoff means SMEs can maintain and tune the workflow themselves through a no-code builder.⁴

Phase 5 — Maintenance & expansion (the land-and-expand)

Once one workflow is in production, the platform license + retainer recur, and the pod expands to adjacent workflows in the same account. This is the Palantir land-and-expand: services-heavy "land," platform-centric "expand." Distyl claims a 100% production record reaching 120M+ end users,¹ CONFIRMED (company) and individual accounts now run many workflows (e.g., the telecom account spans devices, plans, orders, promotions, and new sales on one reasoning layer).¹²

Is "quarters-to-weeks" real?

Partly. The disclosed timelines are aggressive but inconsistent, which tells you the truth: the first wedge can be fast, the full transformation is not.

• Auto-finance lender: "Live in 1 week" from kickoff to agentic fraud detection.¹² CONFIRMED (company)
• Prior-authorization decisioning: "Deployed in 1 quarter."¹¹ CONFIRMED (company)
• Telecom CX and CPG order-resolution: "Deployed in weeks" / "Onboarding in weeks."¹² CONFIRMED (company)

INFERRED: the "1 week" cases are almost certainly the narrow, highly-checkable sub-workflow (document intelligence with a clean system-of-record verifier), while "1 quarter" is the realistic figure for a production decisioning system in a regulated domain. The honest read: weeks-to-first-value is credible; quarters-to-full-rearchitecture is the real arc, and the PR conflates the two.

What "own the outcome" means — contractually and operationally

Operationally: the FDE pod shares accountability — "you need to forward deploy it at the clients to own the outcome with them," and they don't leave after a demo.³ Contractually: "part of Distyl's project fee is tied to achieving a client's objectives," an explicit outcome-based contract versus time-and-materials.³ CONFIRMED (founder interview) The exact structure (what % is at risk, what triggers payment) is not disclosed — the only thing confirmed is that a portion, not all, of the fee is outcome-linked, and that platform-licensing fees recur on top.

3.Team shape & who you hire

The org chart is the cost structure. A founder is mostly buying a payroll of expensive, customer-facing engineers, so the FDE profile, the ratios, and the recruiting pipeline determine whether the unit economics work.

What an FDE actually is (three archetypes)

"Forward-deployed engineer" is used for three different jobs. A deployment startup wants almost entirely the first:⁹

REPORTED — Bloomberry analysis of 1,000 FDE job postings (Revealera data), Jan 2026.⁹

The hire profile

• Seniority: mid-to-senior. Of FDE postings that specify, ~60% want 3–5 years, ~20% want 6–8, ~8% staff+; only ~12% are entry-level.⁹ Distyl's own forward-deployed AI engineer posting asks for 2+ years (production systems) up to 3+ (general AI engineer).⁷
• Stack: Python (66% of postings), TypeScript (35% — they build full-stack dashboards for customers), AWS/GCP/Azure, Docker/Kubernetes.⁹
• AI skills: AI agents (35%), LLMs (31%), RAG (12%) — generative/agentic, not classical ML. Distyl explicitly wants people who design "models, agents, retrieval, and evaluation."^7,9
• Soft skills are the differentiator: able to sit with a non-technical VP, thrive in ambiguity, and own a deployment end-to-end. Distyl's leadership is drawn from Palantir, Apple, Microsoft, OpenAI, Snorkel and national labs.^2,7

Where FDEs sit, and the ratio question

Organizationally, ~45% of postings place FDEs on their own dedicated team and ~38% inside the product-engineering org — not under sales (14%) or customer success (7%).⁹ This matters: a deployment startup is structurally an engineering company with a field arm, not a consultancy with coders.

INFERRED (FDE:platform ratio & pod size): neither company discloses team sizes. From the model's mechanics — small pods embedded per account, feeding a central platform team — a reasonable estimate is 3–8 FDEs per active engagement and an FDE:platform-engineer ratio of roughly 2:1 to 4:1 early, compressing toward 1:1 as the platform absorbs field-learned patterns. Treat these as Galileo estimates, not disclosed figures.

The recruiting pipeline (ex-Palantir + the broader funnel)

The headline edge is the ex-Palantir network: both Distyl founders ran Palantir deployment/BD,³ and Northslope's CEO Bill Ward is a former Palantir FDE who staffed the company from Palantir alumni.⁸ But the broader hiring funnel is wider than just Palantir — analysis of 100 FDE LinkedIn profiles shows most come from software engineering (45%) and solutions/architecture (22%), with consulting (10%) and data (15%) feeding in too. Notably, the same analysis flagged a Northslope FDE who came from McKinsey and a Distyl-adjacent pool from OpenAI/Apple.⁹ The ex-Palantir pipeline is the brand and the early core; it is not the only source of supply.

The labor market a founder is hiring into

The FDE talent market is real but thin, and the comp it commands is the single biggest line on a founder's P&L. The scale signal most often quoted is ~224 open FDE roles across 39 companies — but treat it as a single-source point-in-time snapshot, not corroborated market data: it originates with one tracker (Jobs By Culture, May 30 2026), and the exec-search and Substack pieces that repeat it all trace back to the same count.²¹ REPORTED (single tracker) It directionally agrees with the labs themselves staffing up: a16z's own mid-2025 count found 22 of 311 open OpenAI roles were forward-deployed / solutions engineering.¹⁴ CONFIRMED (a16z count)

On comp, anchor to the hard, official numbers and treat aggregator bands as estimates. The Tier-1 anchors from companies' own careers pages: OpenAI FDE $180K–$280K base + equity²² and Scale AI FDE $179K–$224K base.²³ CONFIRMED (official postings) These bracket Distyl's own $150K–$250K base posting.⁷ The wider, splashier bands — ~$250K–$640K all-in at startups and ~$190K–$420K at F500 buyers — come from recruiter and career-guide aggregates, not primary salary data, and should be read as ESTIMATED directional ranges, not disclosed figures.⁹

4.Tech / architecture & the verification layer

This is where the "genuine IP vs orchestration glue" question gets answered. The short version: the differentiated asset is not the model and not the agent — it's the pipeline that turns messy proprietary context into auditable, verifiable production logic that a regulated enterprise will trust enough to pay outcomes on.

Distyl's Distillery: ingestion → business-logic mapping → agent generation

The clearest public description of how Distillery works comes from Sacra's product teardown, corroborated by Distyl's own platform language.^10,12 The pipeline:

1. Ingest the SOP. A user uploads a written standard operating procedure (e.g., "approve refund requests," "process insurance claims").¹⁰
2. Decompose into discrete tasks ("Routines"). Distillery auto-decomposes the SOP into checkable steps — e.g., validate order number → check warranty status → issue refund — converting prose procedures into auditable AI workflows.¹⁰
3. SME refinement via no-code builder. Subject-matter experts review and refine the auto-generated prompts and tool-chains, attach test cases, and provide feedback examples — without writing code.¹⁰
4. Connect to systems of record. Routines bind to the customer's data via SQL, APIs and document repositories, under role-based access, multi-tenancy, and SOC-2 controls.¹⁰
5. Run with full auditability. In production, the platform captures every input, output, tool call and reasoning step; failed tasks trigger alerts or human-in-the-loop review; execution traces feed feedback that improves the system via prompt adjustments and A/B testing.¹⁰

SiliconANGLE's coverage adds the reliability framing: every workflow is evaluated for performance, reliability and correctness, with observability over every input/output and guardrails against hallucination.¹³ Distyl runs primarily on Microsoft Azure infrastructure and adapts OpenAI and Anthropic frontier models for deployment.^3,4

Northslope's Palantir-native stack

Northslope builds the same kind of mission-specific agentic apps, but on Palantir's substrate: AIP, Foundry and Gotham, with Google Cloud/Gemini.^2,8 Its differentiator is "ontology-first AI" — building applications around Palantir's structured, machine-readable model of the enterprise's entities and business logic, rather than around a black-box model.⁸ Ward's pitch: "Ontology-first AI is the only option for enterprises who aim to lead their industries for the next 100 years."⁸ The architectural trade is explicit — Northslope inherits Palantir's deterministic ontology and governance for free, but pays Palantir the substrate rent and is locked to that ecosystem.

The verification layer — why outcomes are billable at all

You can only sell an outcome if you can verify it. The intellectual spine here (from prior primary-source work, grounded in DeepMind's "Large Language Monkeys," 2024¹⁵): generating candidate answers is cheap and commoditizing; the verifier that selects the correct candidate is the scarce asset.

The DeepMind result: coverage (does any of N samples solve it?) scales log-linearly across four orders of magnitude of sample count. DeepSeek-Coder-V2 on SWE-bench Lite went from 15.9% @1 sample to 56% @250 samples, beating the 43% single-sample SOTA — a weaker model sampled often beats a stronger model sampled once.¹⁵ REPORTED (peer-reviewed preprint)

This is exactly the design visible in Distyl's case studies: the verifiable wedges (document extraction with consistency checks, contract extraction against source docs, decisioning with citation trails) are precisely the tasks where an objective verifier already sits in the customer's stack.¹²

Genuine IP vs orchestration glue — the honest split

5.Business model & unit economics

Three revenue legs, one motion. The trick a founder needs to understand: the business is sold as services (low margin) but valued as software (high margin), and the bet is that the platform leg grows fast enough to bend the blended margin upward over time. This is a16z's "trading margin for moat" / "services-led growth" thesis applied directly.¹⁴

The three revenue legs

1. Services (FDE engagements) — embedded engineering, billed in part on outcomes. The "land."
2. Platform license — recurring Distillery (Distyl) or Palantir-native app (Northslope) licensing; customers use it to operate/maintain their AI systems at scale. The "expand."^3,10
3. Outcome fees — a portion of fee tied to hitting the client's objective, layered onto services.³

The numbers table

Margins: services vs software (and where these land)

INFERRED. Neither company discloses margins. As industry anchors: pure professional services run ~30–50% gross margin; enterprise software runs ~70–90%. The whole valuation case rests on the blend bending toward software. A useful comparable: Harvey (legal AI, explicitly "software licensing + intensive forward-deployed services") is characterized at 85%+ gross margin despite a service-heavy GTM.¹⁶ REPORTED Distyl's "backed by profitability" claim in year two is unusual and, if true, suggests services are at least covering cost — but no margin figure is public. Galileo's honest estimate: blended gross margin today is likely in the 40–60% band (services-weighted), with the platform leg as the lever to push it higher.

How outcome pricing actually gets measured

The mechanism is: define the success metric during scoping (e.g., resolution-time reduction, decisioning accuracy, cost saved), instrument it in the platform's observability layer, and tie a portion of the fee to hitting it.^3,11 The objective verifier doubles as the billing meter — you can charge on "containment rate" only because the system logs every contained interaction. What's not disclosed: the % of fee at risk, the floor/ceiling structure, or NRR. INFERRED: given deep workflow integration and multi-year contracts, NRR is plausibly well above 100% (land one workflow, expand to many), consistent with the Palantir land-and-expand pattern — but this is a model inference, not a reported number.

6.Sectors, customers & what they automate

Sector concentration — the actual vertical mix

Distyl's disclosed and reported customer base over-indexes on regulated, document-heavy, decision-intensive industries: healthcare/payers, telecommunications, insurance, manufacturing/supply-chain, financial services, CPG — plus federal (HHS noted in prior work).^1,13 The case-study evidence is heaviest in healthcare payers (three of six public cases) and supply-chain/operations. The logic is the verification thesis made concrete: these verticals have systems of record (claims DBs, contracts, ERPs) that act as built-in objective verifiers, and they have expensive, repetitive, auditable workflows where "own the outcome" is provable. The broader FDE market confirms the gravity: financial services (24%), government/defense (18%), healthcare (17%) and insurance (17%) are the top customer industries for FDE roles overall.⁹

Named / credibly-reported logos

• T-Mobile USA — named by The Information / SiliconANGLE as a Distyl customer with value-generating AI projects (plausibly the F100 telecom CX case).¹³ REPORTED
• All six published Distyl case studies are anonymized by tier ("F20 healthcare payor," "F50 hardware manufacturer," "F100 telecom operator," "publicly traded auto lender," "F50 CPG brand").¹² Specific names beyond T-Mobile are not disclosed.
• Northslope: specific logos not disclosed; it reports a "large-scale expansion" deal tied to Palantir AIP/FDE deployment and Vanguard: Elite partner status.⁸

Granular automation case studies (before-state → what the agent does → result)

GTM tactics

• Founder-led, ex-Palantir brand + a board/cap-table stacked with F500 directors and marquee angels (Nat Friedman, Brad Gerstner).¹⁰
• Outcome guarantees as a wedge against incumbent consultancies' time-and-materials.³
• Lab + functional partnerships — Distyl partners with OpenAI and Anthropic for models, Azure for infra, and "functional-expertise" firms (clinical ops, insurance ops, supply-chain ops) to run complex programs.³ Northslope's entire GTM is the Palantir partnership (first Vanguard: Elite partner).⁸
• Regulated-industry focus — leaning where auditability and systems-of-record verifiers exist.

Where it stalls or breaks

• Fuzzy, no-verifier work. Open-ended strategy, design, taste — no objective oracle exists, so the weak LLM-judge path is the only verifier and outcome billing becomes a liability. Don't sell magic there.
• Dirty data foundations. Prakash names a "shaky data foundation" as a core adoption blocker; without SME involvement, "there's no chance we're able to go into production."³
• Implementation/change-management risk. Sacra flags long sales cycles and execution risk; a failed F500 deployment damages the 100%-production-record brand disproportionately.¹⁰
• Margin drag if the platform leg lags. If services don't convert to license revenue, you're a consultancy at a software multiple.

7.The peer field

These are the independent companies running an embed-and-build-on-customer-data motion adjacent to Distyl/Northslope. For each: what they do, and how similar vs. different to the FDE deployment model.

The pattern across peers: almost everyone has adopted some combination of (a) outcome/KPI pricing, (b) forward-deployed delivery, and (c) "harvest the customer's proprietary context." The differentiators are vertical depth vs. horizontal breadth and product-led vs. FDE-led. Distyl/Northslope sit at the FDE-led, multi-vertical, embed-deep end — the most services-heavy, highest-touch corner of the field.

The smaller-player ladder — who's running this at each funding stage

Beyond the unicorns above, a fast-growing cohort of earlier-stage companies is running the same ex-Palantir embed-and-build playbook, and the most useful comp set for understanding the model's formation is tiered by stage. The recurring pattern is unmistakable: ex-Palantir / ex-FAANG founders → pick one messy vertical workflow as the wedge → embed an FDE to build directly on the customer's data → fold the bespoke builds back into reusable platform infrastructure. The earlier the stage, the more visible the seam between "services to land" and "platform to expand."

The read at seed: the wedge is everything — nobody wins horizontally this early, and the defensible ones go vertical-deep first (CPG ops, insurance docs, compliance, B2B order intake). For a would-be founder the competitive math is sobering: you'd be roughly the fifth-plus ex-operator running the identical move, so the edge has to come from a sharper wedge, proprietary data access, or raw deployment muscle — not from the motion itself, which is now common knowledge.

8.Palantir lineage & the competitive field

Palantir as the FDE ancestor

The FDE model was invented at Palantir in the early 2010s for intelligence customers who couldn't share requirements in advance: instead of "requirements → spec → product," Palantir embedded engineers on-site to observe real workflows and build in situ. Shyam Sankar rebranded these "solutions/integration engineers" as Forward Deployed Engineers (~2011); commercial FDEs were "Deltas," government FDEs "Echoes."¹⁴ The crucial insight a16z draws: Palantir FDEs were not doing consulting — they were extracting how institutions run and encoding it into the product (the Ontology), so every expensive deployment compounded into reusable platform value.¹⁴

Deterministic ontology vs. probabilistic agents

The genuine difference between Palantir-classic and the new AI-FDE wave: Palantir's Ontology is a deterministic, auditable model of entities, relationships and allowed actions — repeatable and traceable for high-stakes domains. The new wave layers probabilistic LLM agents on top. The hard problem the deployment startups solve is exactly the seam: wrapping non-deterministic models in deterministic, evaluable, ontology-or-verifier-constrained operations.¹⁴ Northslope's bet is to keep Palantir's deterministic ontology as the control surface; Distyl's bet is to build its own (Distillery's Routines + verifier layer).

Why this ex-Palantir wave, now

a16z's "Palantirization of everything" / "services-led growth" thesis is now the canonical playbook: opinionated platform + embedded engineers + multi-year outcome-aligned contracts (reaching tens of millions) beat toolkits-plus-generic-consulting.¹⁴ With models commoditizing and AI startups going $0→$20M ARR fast, services velocity (FDE capacity) has to keep up — and the people who already know how to do this are the Palantir diaspora.

That diaspora is now well-documented as a founder factory. The WSJ has profiled the "Palantir mafia" behind a wave of Silicon Valley startups,²⁴ and Business Insider reports that the FDE role — Palantir's "Deltas," the tip-of-the-spear engineers who embed with clients — "churns out startup founders," with one alum calling it "founder preparation bootcamp."²⁵ Named FDE→founder examples include Barry McCardel (Hex), Gary Lin (Explo), Eliot Hodges (Anduin), and the Trae Stephens / Matt Grimm Anduril founding team. BI's own scan found ~700 LinkedIn profiles listing both Palantir and "founder" — REPORTED (BI's LinkedIn count), a headline number that should be read as an order-of-magnitude signal, not a clean census.²⁵

Industry analysts frame why the skillset transfers. Everest Group calls Palantir's FDSE model a "category of one" — a hybrid that "operate[s] like a product company in its platform approach but deploy[s] like a consulting firm in its proximity to the customer," demanding opinionated platforms, embedded execution, and operational credibility at once. Everest's sharpest distinction: a Dev builds one capability for many customers; an FDSE enables many capabilities for a single customer.²⁶ REPORTED (Everest Group) That is exactly the muscle a deployment-startup founder is trying to bottle.

Brief competitive context (≈10%)

Lab arms: OpenAI and Anthropic are building their own forward-deployed/applied-AI teams, and both partner with Distyl on models while also competing for the same enterprise transformation budget.³ The dynamic is partner-and-compete: today the labs supply the substrate; tomorrow their applied arms may go direct. Consulting/SI response: Accenture, Deloitte, Cognizant, PwC — plus federal SIs (Booz Allen, Leidos, SAIC) — have adopted "FDE" labeling, but a TBR analyst notes smaller/nimbler firms have the adoption advantage because large SIs struggle to scale outcome-based commercial models across their workforce.³ The disclosed numbers show the SIs are getting traction even if the motion is hard to scale: Accenture reported FY2025 GenAI new bookings of $5.9B and $2.7B in advanced-AI revenue — CONFIRMED (official filing) — though in context that is only ~7% of total bookings and ~4% of revenue, a reminder that even the largest SI's AI book is still a thin slice of the whole.²⁷ EY's tie-up with 8090 is the clearest example of an SI buying the motion rather than building it: EY named 8090 a founding partner of its EY.ai PDLC, powered by 8090's agentic "Software Factory" (~Mar 2026); note EY's headline 80×-faster / ~70% productivity claims are VENDOR-CLAIMED, not independently benchmarked.²⁸ A different shape entirely is Thrive Holdings, a Thrive Capital–backed permanent-capital roll-up that buys traditional services firms and re-builds them around AI (with an OpenAI equity stake) — a services roll-up, not an FDE integrator.²⁹ Platforms as context, not peers: Scale AI, Databricks and Palantir itself compete for budget but are infrastructure/platform plays, not the independent embed-and-build startup a founder would be standing up.

9.So what for a founder

An honest synthesis for someone deciding whether to build one.

The distinct ways to build one

1. Own the substrate (the Distyl path). Build your own platform (a Distillery-analog). Highest ceiling — platform margin, software multiple, $1.8B-type outcomes — but capital-intensive and slow; you're racing a well-funded incumbent and the labs on the substrate.
2. Rent the substrate (the Northslope path). Build mission-specific apps on Palantir AIP (or a hyperscaler stack). Faster to revenue, lower capital, inherits governance/ontology for free — but you pay the substrate rent and your ceiling is capped by the platform owner's goodwill and economics.
3. Vertical wedge with a built-in verifier. Pick one vertical where (a) an objective verifier already sits in the customer's systems of record (claims DBs, ledgers, code/IaC, reconciliation, document→schema), and (b) you have unfair domain access. This is the most defensible entry for a new team: you don't out-platform Distyl, you out-embed them in a niche where you have the relationships and the verifier is free.

What's commoditizing

• The model — rented, swappable, deflating in price.
• The agent/RAG orchestration — increasingly off-the-shelf.
• The verification harness as a concept — the idea of sample-and-verify is now widely known; the FDE label itself is being commoditized by every consultancy.⁹
• The "we deploy AI for enterprises" horizontal pitch — now a knife fight against Distyl + the labs + the SIs.

Where the open lanes are

The edge moves to one of three places — pick at least two:

• Substrate ownership — if you can actually build and defend a platform layer (hard, expensive, but highest ceiling).
• Proprietary context / data access — privileged access to a customer's or vertical's data and the verifier wired into it; this is sticky and accumulates proprietary pass/fail traces a lab can't replicate.
• Deployment muscle / relationships — the real Palantir moat was never the software, it was being embedded. Credibility and F500/vertical access that lets you land and never get ripped out.

Bottom line. Distyl is the existence proof that this is a venture-scale business; Northslope is the proof you can run the same motion on rented rails one stage behind. The function (embed → harvest context → build agents → bill on outcomes) is commoditizing. The money is in owning the scarce parts the body actually supports: the verifier integrated into the customer's reality, the proprietary traces it generates, and the embedded relationship that makes you impossible to rip out. The model is rented. The verifier and the relationship are owned. That's the company.

§Sources

All citations resolve to real, named, publicly accessible sources accessed June 16, 2026. Anonymized customer case studies are labeled as company-published.

1. Distyl AI, "Distyl AI Raises $175 Million at $1.8 Billion Valuation to Help Global Enterprises Become AI-Native," PRNewswire, Sep 22, 2025. link
2. "Applied AI Company Northslope Raises $22M to Replace One-Size-Fits-All Enterprise SaaS," BusinessWire, Jan 29, 2026 (round details corroborated via Pulse2 & Preqin). link · Pulse2
3. Channel Dive, "Billion-dollar AI startup leans on collaborative deployment model, outcome-based pricing" (Prakash interview), 2025. link
4. Distyl AI platform / company site (Distillery AI Engine description). distyl.ai
5. Fenwick & West, "Fenwick represents Distyl AI in $175M Series B funding at $1.8B valuation." link
6. Distyl AI careers / "Forward-Deployed AI Engineer (all levels)" posting (Ashby). jobs.ashbyhq.com/Distyl
7. Distyl "Forward-Deployed AI Engineer — Production Systems" posting (Ashby; mirror via Vaia listing $150K–$250K base). Ashby · Vaia mirror
8. Northslope Technologies — Bill Ward founder/CEO, ontology-first, ~88 employees, Vanguard:Elite Palantir partner, large-scale expansion (Yahoo Finance/Reuters wire, Pulse2, BriefGlance, LinkedIn). Yahoo/wire · BriefGlance · northslope.com
9. Bloomberry, "I analyzed 1,000 forward-deployed engineer jobs — what I learned" (Revealera data), Jan 2026. link
10. Sacra, "Distyl AI valuation, funding & news" (Distillery product mechanics, contract values, total raised ~$202M, seed angels Friedman/Gerstner). link
11. Distyl AI case study — "F20 Healthcare Payor — Prior Authorization." link
12. Distyl AI case-studies index + individual cases (telecom, auto-finance, CPG, contracts, manufacturer). case-studies · auto-finance · telecom · CPG · contracts
13. SiliconANGLE, "Enterprise AI consultancy Distyl AI's valuation soars to $1.8B after bumper funding round" (T-Mobile, OpenAI partnership, case studies), Sep 22, 2025. link
14. a16z, "The Palantirization of Everything" & "Services-Led Growth" (FDE history, ontology, trading margin for moat). Palantirization · Services-led growth · FDE origin corroborated at fde.academy
15. B. Brown et al. (DeepMind), "Large Language Monkeys: Scaling Inference Compute with Repeated Sampling," 2024 (coverage scaling; DeepSeek-Coder-V2 SWE-bench Lite 15.9%@1 → 56%@250). arXiv:2407.21787. arXiv
16. Harvey AI — funding trajectory ($3B→$5B→$8B→$11B), "software licensing + forward-deployed services," 85%+ GM (Sacra, TechCrunch, Forbes, FourWeekMBA, Harvey blog). Sacra · TechCrunch · Harvey
17. ValueCore / Distyl value-selling doc on the F50 hardware-manufacturer RCA engagement (80% target, related 40% annual cost reduction). link
18. Sierra / Bret Taylor on outcome-based pricing and ~$100M ARR in 7 quarters (Sierra resources; Economist interview; Lenny's). Sierra · Economist
19. Cresta — contact-center deployment lifecycle, "AI-to-test-AI" eval, $280M+ raised (Cresta blog/guides, PRNewswire, CMSWire). Cresta · PRNewswire
20. Glean — $150M Series F at $7.2B (Jun 2025); $200M+ ARR doubled in ~9 months (Glean press, Reuters via TradingView, Futurum). Glean · ARR
21. Jobs By Culture, "The Forward Deployed Engineer Boom 2026" (~224 open FDE roles across 39 companies; snapshot May 30, 2026 — single-tracker point-in-time count). link
22. OpenAI careers — "Forward Deployed Engineer (San Francisco)" posting ($180K–$280K base + equity). link
23. Scale AI careers — "Forward Deployed Engineer, GenAI" posting ($179K–$224K base). link
24. The Wall Street Journal, "The Palantir Mafia Behind Silicon Valley's Hottest Startups." link
25. Business Insider, "Palantir's forward-deployed engineer role churns out startup founders" (Jun 2025; "Deltas," founder quotes McCardel/Lin/Hodges, ~700 LinkedIn Palantir+founder profiles). link
26. Everest Group, "Palantir: Inside the Category of One — Forward Deployed Software Engineers" ("category of one"; Dev vs. FDSE distinction). link
27. Accenture, Q4/Full-Year FY2025 earnings release & FY2025 Letter to Shareholders (GenAI new bookings $5.9B; advanced-AI revenue $2.7B). Accenture Newsroom
28. EY, "Ernst & Young LLP and 8090 launch EY.ai PDLC" (8090 named founding partner; agentic "Software Factory"; ~Mar 2026; 80×-faster / ~70% claims are vendor projections). link
29. The New York Times (DealBook), "OpenAI and Thrive Holdings" (Thrive Capital–backed permanent-capital services roll-up rebuilt around AI, with OpenAI equity stake), Dec 1, 2025. link
30. Sacra, "Cresta AI — valuation, funding & revenue" (~$270–276M total raised; ~$1.6B valuation, Dealroom/GetLatka concur; one source cites $750M — neither a primary filing). link
31. TechCrunch, "Two Palantir veterans come out of stealth with Edra" (Mar 18 2026) — Edra $30M Series A led by Sequoia (with 8VC, A*); founders incl. Palantir's first forward-deployed AI engineer Yannis Karamanlakis and AIP-launch lead Eugen Alpeza. link / edra.ai
32. Y Combinator company directory — Asteroid (W25), Overstand Labs (W25), Zeit AI (S24), Corvera (W26), Comena (S25), Hyperspell (F25), Karumi (F25), FurtherAI; seed/pre-seed figures and team backgrounds per company profiles & Work at a Startup. link
33. Sapphire Ventures — Bretton AI (formerly Greenlite) $75M Series B (Sapphire Ventures lead, Feb 2026); agentic financial-crime compliance (AML/KYC/KYB/sanctions/transaction monitoring). link
34. Air Street Capital — Fern Labs ~$3M pre-seed; horizontal agent orchestration, three ex-Palantir founders (UK). link
35. General Catalyst / press — Eudia ~$105M Series A (General Catalyst lead, 2025); corporate-legal AI agents for in-house legal teams. link
36. Reuters, "HappyRobot raises $44 million to expand AI agents for freight operators" (Sep 3, 2025) — $44M Series B led by Base10 Partners at ~$500M valuation; existing investors a16z, Y Combinator; ~$62M total raised. link