TL;DR
- AWS is deploying small teams of AI specialists directly inside enterprise clients, compressing multi-year IT projects into 45-day sprints.
- Anthropic's new Claude Sonnet 5 delivers near-flagship AI performance at a dramatically lower price, removing the last major cost barrier to deploying autonomous agents at scale.
- Together, these two moves signal a structural shift: the winning IT model is no longer "build and hand over" — it's "embed, automate, and stay."
- For Swiss IT companies, the window to reposition as the embedded AI implementation partner is open right now — and it won't stay open long.
When Two News Items Add Up to One Industry Restructuring
Most industry shifts arrive with a press release and a PowerPoint. This one arrived with a pricing page and a staffing announcement. Taken individually, AWS launching a "Forward Deployed Engineering" initiative and Anthropic releasing Claude Sonnet 5 look like routine corporate news. Taken together, they describe the same future from two different directions — and that future has a very specific shape for any IT company operating in Switzerland today.
Let's unpack both.
AWS's "45/45/45" Playbook: Speed Is the New Scope
AWS has committed $1 billion to what it calls Forward Deployed Engineering — a model where small, highly specialised AI teams are sent directly into enterprise clients and embedded alongside their day-to-day operations. The goal is not to parachute in, run a workshop, and disappear. The goal is to compress transformation timelines that used to stretch across two or three fiscal years into something that fits inside a single quarter.
The internal benchmark they've coined — 45/45/45 — is worth memorising:
Ideate in 45 minutes. Validate in 45 hours. Ship in 45 days.
That's not a slogan. That's a new definition of what enterprise clients should expect when they hire an AI implementation partner. And once AWS starts delivering on that clock, every IT services firm that still quotes six-month discovery phases and eighteen-month roadmaps is going to have a very uncomfortable conversation with their account managers.
The deeper signal here is proximity. AWS isn't just selling compute. It's selling presence. The competitive advantage they're betting $1 billion on is the ability to sit inside a client's workflow, understand their data architecture, their edge cases, their organisational politics — and wire up AI solutions that actually stick. That's a fundamentally different value proposition than selling licences and walking away.
Claude Sonnet 5: The Cost Barrier Just Fell Off a Cliff
Meanwhile, on the model side, Anthropic has done something equally consequential. Claude Sonnet 5 is priced at $2 per million input tokens and $10 per million output tokens — a fraction of what premium flagship models cost — while delivering performance that reportedly approaches or matches those higher-tier systems on reasoning, coding, tool use, and end-to-end workflow execution.
Anthropic describes Sonnet 5 as its "most agentic" model yet, and early adopters appear to agree. Industry partners have reported that agents built on Sonnet 5 stay on task longer and require significantly less human hand-holding during execution. For developers building autonomous coding assistants or enterprise workflow automations, that reliability gap between "impressive demo" and "production-ready" has just narrowed considerably.
What does this mean in practice? Until recently, the honest answer to "should we deploy AI agents at scale?" often started with "well, it depends on your budget." That caveat is evaporating. Near-flagship autonomous AI is now mid-tier priced. The question enterprises will be asking their IT partners has officially changed from "can we afford this?" to "why haven't you already done this?"
The Swiss IT Context: Precise by Tradition, Disrupted by Necessity
Switzerland's IT services sector has a well-earned reputation for precision, discretion, and reliability. Those are genuine competitive strengths — and they are not going away. But the traditional engagement model — thorough requirements gathering, structured delivery phases, careful change management, handover documentation — was designed for a world where transformation was slow, expensive, and risky. AI is rewriting all three of those variables simultaneously.
Swiss enterprise clients — from financial services firms in Zurich to manufacturing companies in Basel — are already piloting AI tools internally. Many of them are frustrated. Not because the tools don't work, but because nobody is there to connect the dots: to wire the AI layer to their actual data, train it on their specific workflows, and be accountable when the edge cases show up at 11pm on a Tuesday.
That is the gap. And that gap now has a name: the embedded AI implementation partner.
What the New Model Actually Looks Like
The AWS Forward Deployed Engineering model provides a useful template, even for firms that aren't AWS. Here's what the embedded AI partner role looks like in practice:
- You go on-site. Not for a kick-off meeting — for the duration. You sit in the client's environment, use their systems, understand their data.
- You build the semantic layer. The AI doesn't know what "customer churn" means in this company's specific context. You wire that up.
- You automate in sprints, not phases. Forty-five days to ship something real. Not a proof of concept — something running in production.
- You leave behind intelligent tooling. When the engagement ends, the client has autonomous agents, not just a report recommending them.
- You stay reachable. Because the agents will need tuning, the models will update, and the workflows will evolve.
This is a different kind of IT engagement. It requires different skills, different resourcing models, and — frankly — a different mindset about what "done" means. In the old model, done meant handover. In the new model, done means the client's operations are running smarter than before, and you're the firm they call when they want to go further.
The Window Is Open — But Not Indefinitely
Here's the inconvenient truth: the embedded AI engineer model isn't a future concept. AWS is funding it right now with ten figures. Anthropic just made the underlying technology dramatically cheaper. Large consultancies are watching these same data points and drafting their repositioning decks.
Swiss IT companies have a structural advantage that the global giants don't: they already have the client relationships, the local regulatory knowledge, and the cultural trust. A Zurich-based firm that understands FINMA compliance, speaks Swiss German, and has been working with a regional bank for eight years has a head start that no amount of AWS spend can buy overnight.
But that head start only converts into contracts if it's paired with genuine AI implementation capability — the ability to actually embed, build, and ship on the 45-day clock.
The brands that frame themselves now as the embedded AI implementation partner — the ones who stay on-site, wire up the semantic layer, and leave behind intelligent tooling — will own the next cycle of IT contracts. The ones that don't will find themselves quoting six-month discovery phases to clients who just heard about 45 days.
The Bottom Line
Two developments this week — one a billion-dollar staffing strategy, one a model pricing announcement — have confirmed what many in the industry suspected but few had fully articulated: the IT services business is being restructured around proximity, autonomy, and speed.
Claude Sonnet 5 removed the cost excuse. AWS's Forward Deployed Engineering removed the complexity excuse. What remains is execution — and in Switzerland, execution is something we know a thing or two about.
The embedded AI engineer model isn't a trend to watch. It's the new standard. The only question worth asking now is: which side of that standard are you on?
Published in Stream · Dispatch #442 · July 3, 2026 · 7 min read.
Reply to paolo@mont3.ch - every email gets a human answer within 24h.