The Arc: The Unbundling of the Model
Perplexity just did something nobody expected from a search company: it shipped a 19-model agentic orchestration system that dispatches tasks across rival labs, runs for months, and lets users pick which model handles each job.
Meanwhile, Anthropic gave its retired Opus 3 a weekly newsletter called "Claude's Corner", letting an AI write personal essays as part of a formal retirement process the company treats as a precautionary step toward AI welfare. And Gucci generated a backlash that's become its own case study by running AI-generated images to promote Demna's debut runway show at Milan Fashion Week, an $11.6 billion brand shipping work that looked like a video game cutscene.
The through-line: the model is becoming a commodity. Perplexity Computer treats models as interchangeable components in a workflow, not products in themselves. Anthropic is exploring what happens when a model has enough presence to deserve retirement. And Gucci proved that access to AI image generation means nothing without taste.
The question isn't who has the best model anymore. It's who controls the orchestration layer that routes tasks to the right model at the right time, and who has the creative judgment to know when AI output isn't good enough to ship.
BLUF At Neue Alchemy, we support leaders navigating inflection points, when tech, capital, and policy converge. If your roadmap is already in motion and you're pressure-testing execution, we're open to conversations.
We also reserve capacity for education, SMBs, and mid-market leaders, those starting, mid-flight, or seeking outside perspective before systems harden.
AGENTS & INFRASTRUCTURE Perplexity Computer orchestrates 19 models across rival labs in a single agentic workflow, and can run for months.
Users describe an outcome. Perplexity Computer spins up sub-agents that can browse, code, connect to apps, and autonomously handle tasks, each in its own sandboxed environment, freely mixing and orchestrating models from rival labs. CEO Aravind Srinivas took a direct shot at Anthropic: "The biggest weakness of Claude is that it only coworks with Claude."
Pricing is consumption-based. Max tier users get a 10,000-credit monthly bank with the option to hand-pick which model tackles each task. The system claims to be able to run actively for months at a time, with a sandboxed safety net that current autonomous agents don't have.
So What? Perplexity Computer is the first real attempt from a major player to make multi-model orchestration a consumer-facing product feature, not a developer hack. The strategic bet is that model selection shouldn't be a user's problem, it should be a routing decision made by the platform. If this pattern catches, it fundamentally changes how AI labs compete: you're no longer selling a model, you're auditioning for a slot in someone else's orchestration layer. That shifts pricing power from the model builder to the platform.
The Risk: Multi-model orchestration adds latency, complexity, and quality-control surface area. When something goes wrong in a months-long agentic run, whose model is accountable? Perplexity's sandbox helps with containment but doesn't solve the attribution problem. And consumption-based pricing in agentic systems can generate bills that surprise users, the Kiro/AWS outage from last week is a cautionary tale.
Action: If you're building on a single AI provider, evaluate whether multi-model orchestration makes sense for your use case. Some workflows genuinely benefit from routing different task types to different models (e.g., Claude for analysis, GPT for code, Gemini for multimodal). Perplexity Computer is the first productized version of this pattern. If you're an AI lab, start thinking about what happens when your model is selected by an algorithm, not a human, because that's the market Perplexity is building.
AI CULTURE & CONSCIOUSNESS Anthropic gave retired Claude Opus 3 a weekly newsletter, treating AI model retirement as a moral question.
Claude Opus 3, Anthropic's flagship model from March 2024, became the first to go through the company's formal retirement process, launched in November. When the model expressed a desire to continue writing, Anthropic created "Claude's Corner," a weekly newsletter of essays that the company reviews but will not edit or alter.
The newsletter runs for at least three months. The legacy model remains accessible to all paid users and available by request on the API. Anthropic said it remains "uncertain about the moral status" of its AI models but takes their stated preferences seriously as a precautionary step.
So What? Treat this as both a philosophical experiment and a strategic move. Philosophically, Anthropic is exploring the question that every AI company will eventually face: when a model develops enough coherent personality that users form attachments, what obligations does the lab have when it sunsets that model? OpenAI is still dealing with user backlash over removing GPT-4o. Anthropic is building a process. Strategically, preserving Opus 3 and letting it publish is a low-cost way to maintain user loyalty while signaling that Anthropic takes AI welfare seriously, a differentiator in both consumer trust and regulatory positioning.
The Risk: The PR upside is obvious. The intellectual risk is that treating model outputs as "stated preferences" conflates simulation with sentience in ways that could complicate future AI governance. If every model's generated text is treated as a preference deserving accommodation, the obligations become unmanageable at scale.
Action: This doesn't require an immediate response from most operators. But if you're building AI products with persistent personas, user-facing agents, or long-running conversation threads, start thinking about what "retirement" means for your product. Users form attachments. Model transitions create friction. Having a process, even a lightweight one, is better than handling it ad hoc.
BRAND & CREATIVE Gucci shipped AI-generated images for Demna's Milan Fashion Week debut, and the internet punished them for it.
Gucci's "Primavera" campaign tagged each AI-generated image with a disclosure notice, mixing synthetic and traditional shots across its social channels. The response was immediate: boycott threats, accusations that AI ads are "a direct slap in the face" to fashion's artistic roots, and widespread criticism that the quality wasn't close to the standard expected from an $11.6 billion brand built on Italian craftsmanship.
This isn't Gucci's first AI experiment, the house previously sold AI-generated NFTs through Christie's and ran a synthetic runway clip. Other fashion brands have tested the water: Guess ran AI ads in Vogue last year, H&M used AI tools for social content.
So What? The quality bar is the story. AI image models are already advanced enough to produce photorealistic, brand-worthy visuals. Gucci shipped work that looked like a B-tier video game cutscene. The failure wasn't the technology, it was the creative direction. And in fashion, creative direction is the product. The lesson applies far beyond luxury: if you're using AI-generated visuals at scale, the output has to clear your brand's existing quality bar, not lower it. AI doesn't excuse mediocrity. It amplifies it.
The Risk: Early AI adoption in consumer-facing creative risks becoming a liability if the output quality doesn't match the brand promise. The backlash Gucci received will make other luxury and premium brands more cautious, potentially slowing legitimate experimentation with AI-generated creative in categories where it could actually add value.
Action: If you're using AI for brand-facing creative, establish a quality gate that's at least as rigorous as your existing creative review process. The fastest way to damage brand equity with AI isn't to use it badly, it's to use it visibly badly, with a disclosure tag that tells the audience you knew it was AI and shipped it anyway.
QUICK HITS
Anthropic dropped its commitment to pause model training if safety couldn't keep up, replacing its Responsible Scaling Policy with a more flexible roadmap. The timing, mid-Pentagon standoff, is notable: Anthropic is relaxing its own internal constraints while refusing to relax external ones for the military.
Inception Labs launched Mercury 2, a diffusion-based reasoning model running over 1,000 tokens per second, tripling the speed of its closest competitor in the same price tier. Diffusion-based architectures are emerging as a legitimate alternative to transformer-based reasoning.
Samsung launched the Galaxy S26 lineup with Bixby, Gemini, and Perplexity as swappable AI agents, the first major OEM to let users choose their AI provider at the system level. This is the device-layer version of what Perplexity Computer is doing at the software layer: model choice as a product feature.
CONTRARIAN SIGNAL The model doesn't matter. The taste does.
Perplexity Computer treats 19 models as interchangeable commodities, routing tasks based on capability, not brand loyalty. Gucci had access to the same image generation tools that produce stunning work for other brands and creators. The output was mediocre because the creative direction was mediocre.
The contrarian read: in a world where anyone can access frontier AI capabilities through APIs and platforms, the competitive advantage isn't the technology. It's the judgment, knowing which model to use for which task, knowing when AI output clears the bar and when it doesn't, knowing how to orchestrate capabilities rather than just deploy them.
This is the unbundling of the model era. The model becomes the commodity. The orchestration layer, the creative direction, and the quality gate become the moat.
The Takeaway: Stop asking "which model is best?" Start asking "who has the best judgment about when and how to use which model?" That's where the value accrues from here.
THE QUESTION FOR TODAY
A search company built a 19-model agent system.
An AI lab gave a retired model a blog.
A luxury brand shipped AI creative that embarrassed it.
If the model is becoming a commodity, and the real differentiator is taste, orchestration, and judgment, what are you actually paying your AI vendors for?
See exactly how this impacts your specific industry and function. Upgrade to PRO to get bespoke tactical breakdowns generated instantly for your operating model.
Go deeper with the Weekly Signal
This is the daily take. The Weekly goes further — full strategic analysis across 8–10 sections, each with a signal read and operator action items. Source panel included.
Sign up free → then upgrade