Bad AI: Perplexity Scrapes

AI IS ACCELERATING - GET READY

Let's help you understand the WTF of AI and how it helps YOU and your organisation grow. Ready, set, agent.

This week:

😱 Bad AI: Perplexing Scrapes

🔥 AI Playbook: Sales & Marketing Agents

💭 AI Enterprise: Anthropic Vs OpenAI

✨ AI Worlds: DeepMind Genie 3

💡 News, Trends & Insights

BAD AI: Perplexing Scrapes

Tut, tut. AI search engine Perplexity got caught red-handed this week hiding their web scraping intentions, while pretending* to be Google Chrome.

Identity crisis? Quite. They got busted bypassing robots.txt - so websites would think they weren't an LLM sucking the very soul out of the internet.

My head hurts, explain like I'm 5. They hid who they were so they could purloin lots of data and make it look like it wasn't them  make all our AI experiences just super, by stealthily sending millions of daily requests to scrape various websites' data.

Presumably they defended this approach eloquently. Yep, when challenged they flat denied it.

What did they say? It wasn't me (actually 'It wasn't ours' but that ruins the next line).

Ah, the shaggy defence. True, but please, stop quoting 90s one hit wonders, ta.

Why? That don't impress me much.

*allegedly

AI Playbook: Sales & Marketing

Your sales and marketing engine isn’t broken, it’s manual, disconnected and quite likely driving you (and your exec) crazy - why does everything take so long?

Marketing teams lose over 5 hours a week per person to slow research, scattered content that could be reused more effectively and campaigns that take longer to brief than execute.

We’ve put together a playbook on where AI agents can help.

You can download it here.

✅ The 5 areas AI agents work best in sales & marketing
✅ The typical tool stack companies use
✅ What a modular AI marketing system looks like

We build AI sales & marketing agents that plug directly into your pipeline. If you need to scale growth, without adding headcount get in touch.

AI In Enterprise: Anthropic Growing Fast

Anthropic has overtaken OpenAI in AI enterprise market adoption - through a safety-first approach and targeted applications, despite costing a premium over other models.

Key points:

• Market share: Claude has 32% of enterprise AI implementations vs OpenAI's 24%
• Cost surprise: Builders move quickly to new AI models - from the same vendor - even when older models become significantly cheaper.
• Google enterprise push: Gemini climbing to 18% market share through Cloud bundling and existing enterprise relationships.
• Open source decline: 13% of enterprise AI workloads use open-source models, down from 19% six months ago as companies prioritise support and straightforward implementation.

What it means for businesses:

Enterprise AI procurement follows different rules than consumer adoption. CIOs care about predictable behaviour - even if it's not the most cost-effective option.

Wider context:

This mirrors enterprise software - the consumer winner rarely dominates enterprise. Oracle beat flashier databases, Salesforce beat prettier CRMs. Anthropic's focus on 'constitutional AI' (safety) and killer applications like Claude code (also eating up market share) have driven adoption.

Don't count out OpenAI just yet, they still have the top overall mindshare - and we see high usage from companies (Note: This was a small sample of 150 technical decision-makers from startups with $5M+ raised and enterprises with over 5,000 employees). It's still impressive.

AI WORLDS: DeepMind Genie 3

DeepMind has launched a preview of Genie 3, their first real-time interactive world model that generates 3D environments from text prompts - positioning it as a stepping stone toward 'AGI' through its ability to train AI agents.

Key features:

• Real-time generation: Creates interactive environments at 24fps for several minutes (Genie 2 only does seconds).
• Physical consistency: The model remembers what it generated, maintaining world coherence even when revisiting locations.
• Promptable events: Can alter generated worlds mid-interaction - change weather, spawn objects, modify environments through text commands.
• Training acceleration: Enables potentially unlimited scenario generation for agent training without real-world data collection costs or safety constraints.

What it means for businesses:

The training data bottleneck for AI agents just got smaller. Instead of needing massive real-world datasets, this is the beginning of infinite training - and matters to any AI that needs to understand physical environments.

Wider context:

Google is building the infrastructure for embodied AI - agents that can act in physical/virtual worlds rather than just chat. Think customer service avatars that need to understand spatial contexts to AI assistants that help with real-world planning.

Talk to us about AI agent development, we help organisations build effective AI workflows.

WHAT'S NEW - News, Insights & Trends

> Microsoft says these jobs are dead

> OpenAI has new open source models

> Apple wakes up to chat

> Claude cuts off ChatGPT

> The end of gibberish text?

FAQ: What did Perplexity AI do wrong with web scraping?

Perplexity AI was caught bypassing robots.txt files by disguising their web scraping bot as Google Chrome. This allowed them to access and scrape website content that was specifically blocked from AI training purposes, while initially denying the allegations when confronted.

FAQ: How much time do marketing teams waste on manual processes?

Marketing teams lose over 5 hours per week per person due to slow research, scattered content that could be reused more effectively, and campaigns that take longer to brief than execute. This inefficiency is driving adoption of AI sales and marketing agents.

FAQ: Why is Anthropic beating OpenAI in enterprise adoption?

Anthropic has captured 32% of enterprise AI implementations (in a small 150-person study) versus OpenAI's 24% through their safety-first 'constitutional AI' approach, predictable behaviour that appeals to CIOs, and killer applications like Claude Code, despite costing more than competing models.

FAQ: What makes DeepMind Genie 3 different from previous world models?

Genie 3 generates interactive 3D environments in real-time at 24fps for several minutes (versus seconds in Genie 2), maintains physical consistency when revisiting locations, and allows mid-interaction modifications through text prompts like changing weather or spawning objects.

FAQ: What are the 5 key areas where AI agents work best in sales and marketing?

AI agents excel in lead qualification and scoring, content creation and personalisation, campaign optimisation and A/B testing, customer research and competitive analysis, and sales process automation including follow-up sequences and proposal generation.

FAQ: How do enterprise AI purchasing decisions differ from consumer choices?

Enterprise AI procurement prioritises predictable behavior, safety guarantees, and support quality over cost-effectiveness. CIOs value constitutional AI approaches and proven enterprise integrations, explaining why Anthropic's premium pricing hasn't hindered adoption versus cheaper alternatives.