The FDE Deployment Revolution Goes Hyperscale Building on my earlier post about OpenAI and Anthropic launching dedicated Forward-Deployed Engineer initiatives, AWS has now taken a major step of its own. On June 30, AWS announced a $1 billion investment in a new Forward Deployed Engineering organization. The unit will seed thousands of engineers into small "pizza teams" — pods of roughly 5–6 people — who embed directly inside customer organizations to co-develop and deploy production AI systems, with a clear mandate to leave the customer self-sufficient when the engagement ends. This is more than an expansion of professional services. AWS is applying its long-standing "two-pizza team" philosophy — teams small enough to stay agile, autonomous, and fully accountable — to the challenge of enterprise AI deployment. The approach is explicitly "agentic-first," combining human expertise with purpose-built AI agents to compress timelines that traditionally took months into days or short sprints. What makes this significant is the pattern it completes. Palantir pioneered the Forward-Deployed Engineer model over a decade ago. OpenAI and Anthropic recently adapted it with dedicated, well-capitalized deployment entities. Now AWS — the largest cloud provider — is institutionalizing the model at hyperscale from within its core infrastructure business. The message for enterprise leaders is becoming hard to ignore. The primary bottleneck in AI adoption is no longer access to powerful models or abundant compute. It is contextual deployment — integrating these technologies into real workflows, governance frameworks, data environments, and organizational realities. The organizations that will create lasting advantage are those that effectively partner with these teams, internalize the knowledge being transferred, and build genuine internal capability rather than long-term dependency. Full article: https://kentkaufman.substack.com/p/the-fde-deployment-revolution-goes How are you thinking about preparing your teams to work alongside external embedded AI talent? What capabilities are you focusing on building internally?

When your customer research keeps contradicting itself, the problem usually isn't the research. It's that you're talking to two different people and treating them as one. I see this constantly. A founder runs a dozen conversations, and the signal is all over the map. One group swears the onboarding is the dealbreaker; another barely notices it and cares only about price. Half describe the product as a time-saver; half call it a status thing. The founder concludes "customers are confused" or "the market is fragmented," and goes hunting for a cleverer message to paper over the noise. But the noise is the finding. Contradictory insights are what a blended audience sounds like. You're not hearing one confused market — you're hearing two or three coherent ones averaged into mush. And you can't message an average. Nobody is the average customer. The reason this stalls companies is subtle: an average feels like data. It has quotes, a deck, a ring of rigor. So you make product and go-to-market bets on it — and the bets keep missing, because you're aiming at a person who doesn't exist. The fix is in how you structure discovery, before you ever synthesize: Segment on the way in, not on the way out. Decide the cuts that might explain different behavior — job to be done, trigger event, stage, buying context — and tag every conversation as you go. If you only segment after the fact, you've already blended. Recruit on purpose, not on availability. A convenient sample is a blended sample. Deliberately fill each segment so a pattern in one group can't hide inside the average of all of them. Read patterns within a segment, never across the pile. The question isn't "what did customers say?" It's "did this segment say the same thing to each other?" Agreement inside a clean segment is a signal. Agreement across a blended pile is a coincidence you'll pay for later. Here's the bar I hold: an insight is decision-grade only when it's consistent inside a segment you could actually go acquire. Anything short of that isn't evidence — it's an anecdote wearing a percentage sign. The uncomfortable part is that this makes your findings messier before it makes them clearer. Two sharp, opposing truths are worth more than one blurry consensus. Consensus is comfortable. It's also where stalled companies go to feel productive. So a question worth sitting with this week: The last time your customer insights contradicted each other — did you treat it as noise to resolve, or as a sign you were listening to more than one audience at once?

A New Chapter in Enterprise AI Adoption: Learning from Palantir This month, the AI industry took a decisive step from research labs into the heart of the global economy. OpenAI and Anthropic both launched major new initiatives focused on real-world deployment, signaling that the era of AI services at scale has officially begun. Anthropic formed a $1.5 billion joint venture with Goldman Sachs, Blackstone, and Hellman & Friedman to create a dedicated enterprise AI services company. At the same time, OpenAI announced its Deployment Company, backed by over $4 billion in initial capital, and immediately acquired Tomoro, an applied AI consulting firm, bringing roughly 150 experienced Forward-Deployed Engineers (FDEs) on day one. These specialists will embed directly with clients to integrate today's frontier models into complex business operations. This approach is not new. It is a proven playbook pioneered by Palantir. Over a decade ago, Palantir created the Forward-Deployed Engineer model: elite engineers who do not just advise clients but live with them, write production code on-site, customize systems in real time, and own outcomes. This last-mile commitment turned Palantir from a promising data platform into an indispensable partner for the world's most demanding organizations. OpenAI and Anthropic are now adapting this model, but with a critical difference. They are deliberately keeping these enterprise services in separate, heavily capitalized entities rather than folding them into their core companies. Why the separation? Two reasons stand out. First, protecting the AGI mission. Both organizations were founded to push toward safe artificial general intelligence. Their top researchers need to remain focused on breakthrough model development, not on enterprise customization, sales cycles, or compliance work. Second, cultural and operational fit. Deployment is capital-intensive, people-heavy, and involves long sales processes. Spinning out these arms allows each company to attract dedicated capital and talent without diluting the core labs or exposing them to heavier regulatory and liability risks. Palantir showed how to cross the deployment chasm. OpenAI and Anthropic are proving they learned the lesson while fiercely guarding their original purpose. The labs will keep racing toward AGI. The new deployment companies will make today's powerful models essential infrastructure inside Fortune 500 companies, banks, hospitals, and factories. For business leaders, the critical question is no longer about access to technology. It is about whether your organization has the leadership skills to define new requirements, drive real change, and close the gap between advanced models and operational reality. If you are leading AI transformation inside your organization, send us a message ,org and we can bring our playbooks and cirriculumn to help. aiisv.org

Radi8 Social