February 20-27, 2025

For a while, AI felt like a collection of flashy demos—cool and impressive but not always useful in practice. That’s starting to change. The latest wave of AI tools isn’t just about what they can do but how they fit into real workflows.

Claude 3.7 is a prime example. Instead of just answering questions, it adjusts how it thinks based on your needs—quick and snappy when you need speed, methodical when you need depth. Google’s PaliGemma 2 Mix is streamlining visual AI, replacing a mess of specialized tools with one model that can caption, analyze, and even read text in images. And Perplexity’s Deep Research is turning AI into something closer to a true research assistant, capable of sorting through the noise and delivering real insights.

The shift here is subtle but massive: AI isn’t just helping us work faster; it’s starting to change the nature of work itself. Instead of spending time gathering and organizing information, we’ll spend more time deciding what to do with it. Instead of manually optimizing content for SEO, we might soon be optimizing for AI assistants that act as decision-makers for consumers.

That last part is worth thinking about. If AI models are becoming the new gatekeepers—deciding which products, services, and brands to surface—how do we ensure they actually see us?

Let’s get into it.

Anthropic’s latest AI, Claude 3.7 “Sonnet,” is being hailed as the first hybrid reasoning model that can toggle between near-instant answers and extended, step-by-step analysis. In practice, this means Claude can swiftly handle simple prompts or “think longer” on complex ones, revealing its reasoning chain to the user.

Early tests show dramatic gains in coding and problem-solving, but Anthropic deliberately trained Claude 3.7 on real-world business tasks rather than just puzzles. The result is an AI that feels equally at home debugging code or brainstorming a marketing plan. Executives can use Claude’s extended thinking mode for nuanced tasks like campaign strategy or in-depth data analysis, then switch to the fast mode for quick copy tweaks. Notably, Claude 3.7 is widely available (including via API and cloud platforms) at the same cost as previous versions, lowering the barrier for enterprise use.

Alongside the model, Anthropic introduced Claude Code, a command-line AI that can read, write, and even execute code autonomously. While aimed at developers, it’s a glimpse of AI “agents” that could one day handle repetitive digital tasks—imagine a future AI assistant that updates your CRM or generates reports on command.

Claude 3.7’s dual approach marries speed with strategy, which could help teams automate routine work and tackle complex projects with one AI partner. It also increases the competitive pressure on OpenAI and others, as Anthropic leapfrogs into advanced reasoning capabilities before GPT -5’s expected similar features.

Google unveiled PaliGemma 2 Mix, a next-generation vision-language AI that acts as a multi-tool for images. This model can handle multiple tasks with a single system: generating both short and detailed image captions, reading text in images (OCR), answering questions about pictures, and detecting or segmenting objects—all without retraining. It’s essentially a versatile visual assistant: feed it a product image and it can describe it in fluent prose, identify all the items in it, or pull out embedded text.

A marketing team could use PaliGemma 2 to auto-generate rich alt-text and SEO descriptions for hundreds of product photos in seconds or to analyze user-generated images (say, all those customer Instagram posts featuring your product) for insights. Sales and support teams might leverage their OCR to extract info from screenshots and PDFs sent by clients quickly. Because it’s open-source and fine-tunable, one team and one model could replace a tangle of specialized AI tools.

PaliGemma 2 Mix illustrates the trend of consolidating AI capabilities—instead of one model for captions and another for image search, we get an all-in-one visual AI, which can streamline creative workflows and ensure consistency across tasks.

For anyone drowning in information, Perplexity AI has rolled out a new “Deep Research” mode designed to do in minutes what an analyst might spend hours on. The tool performs iterative web searches, reads dozens of sources, and synthesizes the findings into a structured report. It’s like having a diligent research assistant who never sleeps: you enter a broad query (“What are the latest trends in B2B SaaS marketing?”) and Deep Research will scour the internet, then present a concise summary with references. Unlike a standard chatbot answer, it follows a methodical approach—refining its search based on what it finds, and ensuring the final report is coherent and comprehensive. Early reports show it delivers high factual accuracy (93.9% on one benchmark), addressing a common pain point with AI answers.

Competitive analysis, market research, and even drafting whitepapers can be sped up dramatically. Instead of manually gathering data from disparate articles and reports, a strategist could let the AI compile a briefing on, say, emerging consumer behaviors in a region. Perplexity’s tool even allows exporting the AI-generated report to a document or shareable webpage, so the insights can be easily circulated among teams or turned into client-facing material. It’s a freemium product (with advanced features likely behind a paywall), but even the free tier could save teams countless hours.

This reflects a broader move toward AI as a research partner, not just a chatbot. For businesses, it means faster intelligence gathering and the ability to validate assumptions with data quickly. The trade-off is trusting an AI to do the legwork, so due diligence (spot-checking sources) remains key. If widely adopted, tools like Deep Research could change how we prepare for pitches, strategy sessions, and content creation – turning what used to be a labor-intensive process into an almost push-button convenience.

In a milestone for mainstream media, The New York Times has officially embraced AI assistance in its newsroom workflows. This past week, the Times told staff that it is rolling out internal AI tools and training across editorial and product teams. One new tool, an in-house system called “Echo,” can summarize articles and create briefs for reporters. Approved generative AI will also help write social media copy, suggest SEO-friendly headlines, and even generate ideas or questions for interviews.

Essentially, tasks that are time-consuming but not the core creative act of reporting are being offloaded to algorithms. However, the Times is drawing clear lines: journalists cannot use AI to actually write or heavily edit stories or to handle sensitive or copyrighted materials. The move comes with detailed do’s and dont’s to safeguard accuracy and ethics.

The Times’ cautious greenlighting of AI strongly signals that even the most prestigious content producers see these tools as integral to staying competitive (and efficient). For marketing and communications leaders, it’s a case study of adopting AI responsibly at scale. Much like the Times, companies can reap efficiency gains (automatic summaries, content drafts, data analysis) while setting guardrails (e.g., “AI may draft, but a human must approve and polish”). The fact that the Times built its own AI solutions in-house (rather than relying solely on third-party chatbots) also underscores a trend: organizations want custom AI that learns from their proprietary data and adheres to their standards. Executives in other industries might take note that AI adoption doesn’t have to mean sacrificing control or uniqueness.

A provocative idea is taking hold in forward-thinking marketing circles: the next big “customer” you cater to might be an AI. In other words, as consumers lean on AI assistants and recommendation algorithms to make decisions, winning the algorithm could be as crucial as winning over individual humans.

An MIT Technology Review piece this week put it plainly: “Your most important customer may be AI.” What does that mean in practice? Consider how search engine optimization (SEO) forced companies to craft content for Google’s algorithms. Now, instead of just optimizing for human eyeballs, brands must consider how AI models perceive and present their products. For example, if a shopper asks a voice assistant or a chat AI, “What running shoes should I buy?”, the AI will sift through its training data or connected databases to recommend a product. Will it mention your brand? That may depend on what the AI believes about your products (gleaned from descriptions, reviews, social media chatter, etc.).

Startups like Jellyfish already offer tools (e.g., “Share of Model” software) to help brands understand and improve their representation in AI outputs. This can include ensuring your product data is AI-readable, feeding AIs accurate and positive information, and even treating AI responses as a new feedback loop to monitor. It’s akin to SEO, but for AI agents – call it “AI Engine Optimization.” The implications are huge: companies might begin tailoring campaigns or messaging specifically to influence AI recommendations, much as they do for search rankings. If an AI agent ordering supplies for a business always picks the most “AI-visible” vendor, sellers will race to be the one the algorithms favor.

This nascent trend could reshape marketing and sales strategies in the next few years. Executives should start asking: How does our brand appear to AI? Do we need to curate information for AI consumption? Are there partnerships to be made with AI platforms for better placement? As AI intermediaries become gatekeepers to customers, a new form of B2B2C (business-to-bot-to-consumer) marketing may emerge.

We’ll explore this topic in depth in an upcoming special edition, where we’ll share strategies for staying ahead of the curve in appealing not just to hearts and minds but also to algorithms.

— Peter and Torsten