In partnership with
"It's not what happens to you, but how you react to it that matters."
Date: {{current_date_full_with_day}}
Hey {{first_name | AI enthusiast}},
Three signals this week show where AI is really heading.
First: OpenAI is back on top.
GPT-5.4 now ties for the highest score on the Artificial Analysis Intelligence Index. The model pushes ahead in reasoning, coding, and long-context analysis, with a context window exceeding one million tokens.
Second: a personal experiment.
Over the past ten days, a multi-agent system built on OpenClaw is running inside my consulting workflow. Think of it as a junior AI consultant.
One coordinator agent manages specialized agents for writing, coding, research, and illustration. Small AI teams like this may soon become normal.
Third: AI is reshaping commerce, but not the way people expected.
People now use ChatGPT the way they once used Google. They research products, compare options, and narrow choices. But when it is time to pay, they leave.
Finally, a new report from Anthropic offers a reality check. Despite the hype, we are using only a small fraction of AI’s potential at work today. The gap between what AI can do and what people actually do with it remains massive.
Which means the real transformation is still ahead. This edition breaks down what these signals mean for founders, operators, and knowledge workers navigating the AI shift.
I hope you enjoy reading this edition!
PS: If you want to unleash the power of AI agents to grow your business, setup time speak to with our AI experts, here»
GPT-5.4 pushes OpenAI back to the top of the AI intelligence rankings
OpenAI released GPT-5.4, its first major general reasoning model since GPT-5.2. Early benchmarks placed it tied for the most intelligent AI model tested so far.
Key insights
1. Back at the top of the AI leaderboard
GPT-5.4 ranked equal first on the Artificial Analysis Intelligence Index with a score of 57.
That placed it alongside Gemini 3.1 Pro Preview and marked a six-point jump from GPT-5.2, which scored 51.
2. Strong performance in science and coding
The model showed notable gains in scientific reasoning and agentic coding.
In research-level physics tests, GPT-5.4 scored 20 percent, ahead of competing models.
On complex terminal and coding tasks, it achieved 58 percent, again leading the benchmark set.
3. A massive context window
GPT-5.4 expanded the context window to about 1.05 million tokens.
That is more than double the 400,000-token limit in GPT-5.2 and allows the model to process far larger documents or conversations.
4. More knowledge, but also more hallucinations
The model answered more questions correctly than GPT-5.2.
Accuracy improved from 44 percent to 50 percent on the AA-Omniscience benchmark.
However, the model attempted far more answers; this pushed the hallucination rate higher.
5. Large gains in agentic capabilities
GPT-5.4 achieved the highest GDPval-AA score recorded so far at 1,667.
This benchmark measures real-world agentic capability; essentially how well a model can act autonomously across tasks.
6. Better performance, higher cost
Running the full intelligence benchmark cost roughly $2,951 for GPT-5.4.
That is about 28 percent higher than GPT-5.2.
The increase comes from higher per-token pricing despite some efficiency improvements.
So what?
GPT-5.4 shows clear technical progress: stronger reasoning, better coding performance, and a huge context window. But the release also highlights a growing trend in frontier AI; each jump in capability increasingly comes with a higher price tag.
OpenClaw Unleashed: Building an AI Consultant with Multi-Agent Architecture
After 10 days running OpenClaw, I've built a multi-agent system for my consulting business. It functions like an AI Employee for me (Like a junior consultant). But it is getting smarter after every day of onboarding - treat it like a new employee!
The learning curve was steep: I wasn't familiar with terminal setup, but tutorial videos got me started.
The System Structure
A main coordinator agent (Jarvis) oversees three specialized sub-agents (in Discord):
Lois (Writer): Claude Sonnet
Cyborg (Coder): OpenAI's Codex
Oracle (Researcher): Gemini Flash
Picasso (Illustrator): Nano Banana Pro
Maintenance jobs: Haiku (cost-optimized)
Jarvis uses Gemini Flash.
Current Workflow
Jarvis delivers a daily business idea, conducts research to populate my content calendar, and on request, directs Cyborg to build out promising concepts.
Security Approach
The system remains isolated- no external access, email integration, or financial data access. It's functional but it has issues that I solve daily, so I'm treating it as a controlled experiment for now.
Next Steps
Testing continues. Updates to follow as the system matures.
Need help with setup? Get in touch by replying to this email.
My guide to OpenClaw installation is attached here»
ChatGPT is the New Product Research Engine; But Nobody buys anything inside it!
Many people started using ChatGPT the same way they once used Google; ask a question, compare products, narrow options. But when it came time to pay, most users still left the chatbot and bought somewhere else. That behavior forced OpenAI to rethink an ambitious plan to turn ChatGPT into a full shopping platform.
Key insights
🛒 Users researched inside ChatGPT, but rarely bought there
People increasingly used ChatGPT to explore products and compare options. However, they usually completed purchases on traditional platforms instead of inside the chatbot. The gap between research and transaction became clear once OpenAI tested direct checkout features.
🔌 OpenAI shifted from in-chat checkout to partner apps
OpenAI stepped back from letting people pay directly inside ChatGPT. Instead, purchases moved through integrated partner apps such as Instacart, Target, Expedia, and Booking.com. Users could link existing accounts and complete payments through those services.
🏪 Merchant adoption remained extremely limited
The plan also struggled on the supply side. Only about a dozen Shopify merchants were reportedly selling through AI shopping tools, despite millions using the platform. That low participation slowed the rollout of the commerce system.
⚙️ Checkout integration proved harder than expected
Onboarding retailers required manual setup. The company also had not yet built systems to handle tasks such as collecting and remitting sales taxes. These operational hurdles further delayed the strategy.
💰 The shift increased pressure on OpenAI’s revenue model
OpenAI had expected to earn commissions from direct purchases inside ChatGPT. With transactions now routed through partner apps, some of that potential revenue shifted away from the company. At the same time, only around five percent of ChatGPT users pay for subscriptions, which leaves the company searching for additional income sources.
So what?
The experiment revealed something important about AI behavior. Chatbots already influence buying decisions; they guide research, narrow choices, and shape intent. But trust still breaks at the moment of payment.
For now, AI acts like the smartest shopping assistant in the room. The checkout counter still lives somewhere else.
Anthropic Releases its Research report on Jobs
Anthropic released a research report that makes you pause.
Anthropic builds Claude; one of the most advanced AI assistants today. The company was founded by former OpenAI researchers and now sits alongside OpenAI and Google as a major generative AI player.
What makes this study interesting is the data.
Instead of theory, the researchers analyzed millions of real Claude conversations to see how AI is actually used at work.
The result is striking.
The chart compares two things.
Blue shows the work AI could theoretically automate. Red shows what people actually use AI for.
The gap is massive. We are using maybe five percent of what the technology could already handle. Even the most AI-heavy professions are far from saturated.
Computer programming shows about 75 percent coverage.
Customer service sits near 70 percent.
Most other roles barely register.
The biggest exposure is not in low-wage work. It is in knowledge jobs. These roles tend to be older, more educated, higher paid, and often female. The work involves writing, analysis, and decision support.
There are also early signals in hiring. Hiring of workers aged 22 to 25 into AI-exposed roles has dropped about 14 percent since ChatGPT launched.
At the same time, about 30 percent of workers have almost no AI exposure.
Cooks, bartenders, mechanics, and construction workers appear far more resistant to automation.
The takeaway is “simple”. The winners will not compete with AI on routine tasks.
They will focus on what becomes more valuable when intelligence becomes abundant.
Emotional intelligence. Working with AI.
Managing AI agents. Real-world skills.
Close faster with confidence.
Manual tax processes drain your team's time and energy, leaving no time for analysis.
When you automate and centralize your tax data your team can access what they need, when they need it and confidently make decisions.
The best part? Your team gets time back to focus on strategic initiatives.
References
Anthropic research on AI and labor market impacts
Anthropic analyzed millions of real Claude conversations to understand how AI is used at work. The study found AI currently augments human work more than it replaces jobs and is concentrated in knowledge tasks like writing, research, and coding.
Why ChatGPT users research products but rarely buy inside the app
Users frequently use ChatGPT to research products, but few complete purchases directly inside the platform. Limited merchant participation and low buyer conversion have pushed OpenAI to rethink its commerce strategy and rely more on external retail platforms.