Zymbos Intelligence — Newsletter 005: Tools That Compound

Commerce

Shopify opens the ChatGPT front door for 5.6 million merchants - by default

On 24 March, Shopify Agentic Storefronts went live, making every eligible merchant discoverable inside ChatGPT, Microsoft Copilot, Google AI Mode, and the Gemini app - all managed from a single Shopify Admin dashboard. No separate integrations. No additional fees beyond standard processing. The rollout was opt-out by default: millions of merchants were already participating before they read the announcement email. AI-referred traffic to Shopify stores is up 7x since January 2025. AI-attributed orders are up 11x over the same period.

Workforce

Jack Dorsey puts it in writing: AI eliminated 4,000 Block employees

Block - the fintech company behind Square and Cash App - reduced its workforce from approximately 10,000 to fewer than 6,000 employees in early March 2026, the largest single workforce reduction explicitly attributed to AI automation in corporate history. CEO Jack Dorsey's public memo was unusually direct: the cuts were not driven by financial difficulty, but by the growing capability of AI tools to perform a wider range of tasks. Customer support roles were most heavily affected. The company simultaneously announced around 800 new hires focused on AI engineering and machine learning operations.

Product

Sora is gone. OpenAI shut its AI video generator and redirected the compute to robotics

OpenAI has confirmed the shutdown of Sora, its AI video generation tool, rerouting the freed compute toward robotics research. The tool launched in early 2024, briefly topped the App Store following its Sora 2 release, generated significant media coverage, and is now discontinued. It is a clean real-world example of a high-profile AI product with an expiry date: impressive in the moment, without the compounding utility that justifies long-term investment.

Tools

Perplexity ships Custom Skills and an upgraded memory engine - research that learns your workflow

Custom Skills, which shipped on 6 March for Pro and Max users, allow users to save repeatable research workflows and trigger them on demand - no retyping the same prompt every session. Alongside this, Perplexity's memory engine was significantly upgraded: it now recalls important information in 95% of cases, up from 77%, while generating half as many individual memories. The result is a tool that builds an accurate working picture of how you operate and applies it automatically.

Enterprise

Oracle sets aside $2.1bn for restructuring as the people-for-infrastructure trade spreads across Big Tech

Oracle's planned cuts are not driven by falling revenue or missed targets. The company's remaining performance obligations stood at $523bn last quarter, up 433% year on year. The restructuring - which could affect up to 30,000 employees and is funded by a $50bn AI data centre commitment - is a direct trade: people for infrastructure. Analysts estimate the cuts would free up $8bn to $10bn in cash flow. Amazon, Google, Meta, and Microsoft are running the same calculation at larger scale.

UK  ·  Skills

Only 1 in 39 UK job listings requires AI skills — and most that mention it aren't really asking for it

A Prince AI Training study published on 24 March analysed 1,019 non-technical UK job listings across finance, HR, marketing, sales, legal, and operations — the everyday roles that keep British businesses running. Just 2.6% listed AI skills as a requirement. Traditional tools like Excel and Word appeared 7.5 times more often. ChatGPT appeared in 0.6% of listings. Of the 83 roles that did reference AI, more than half mentioned it only in passing — typically describing the employer's product, not a skill expected of the candidate.

This Week's Analysis

The Two Types of AI Tool. Only One Builds an Advantage.

Every AI tool in your stack falls into one of two categories. Most professionals have never thought about which type they are mostly using. That gap is where the real productivity advantage is hiding.

Tools that compound build value over time. They accumulate context about you, your work, and your patterns. Each session adds something to the next. The output you get on day ninety is better than the output you got on day one - not because the tool changed, but because it has learned how you operate. Custom Skills in Perplexity are a simple example. Claude's memory is another. A well-maintained prompt library that reflects months of refinement is a third.

Tools that expire deliver a single output with no carryover. You get the result, you close the tab, the advantage resets to zero. These tools are not useless - they can be highly effective in the moment. But they do not accumulate. Your competitor who uses the same tool gets the same output. There is no moat.

Sora is the clearest recent example of the expiring category taken to its logical conclusion. Impressive, widely covered, and now shut down. The compute has moved because the value was in the moment, not in what built over time.

Run this audit on your stack

Take every AI tool you use regularly and ask three questions:

The goal is not to eliminate expiring tools. Some are genuinely useful in the moment. The goal is to make sure at least part of your stack is compounding - and that you are investing enough time in those tools to let the advantage build.

Perplexity started as a search engine that cited its sources. In March 2026 it has become something more ambitious: an AI research platform that orchestrates multiple frontier models, remembers how you work, and automates repeatable research tasks. The question is not whether it is useful - it clearly is. The question is whether it compounds. The answer is yes, provided you invest in setting it up properly.

Ready to Deploy

The AI Stack Audit

Use in Claude, ChatGPT, or Perplexity. Works best when you list your actual tools.

The conversation about AI in the workplace is dominated by two narratives. The first is productivity: look at everything you can do faster. The second is disruption: look at the jobs that are disappearing. Both are real. Neither is the most important one.

The more important narrative is about accumulation. The professionals and organisations that will define their fields over the next decade are not the ones who adopted AI first. They are the ones who built something durable with it - a stack of tools that compounds, a set of habits that deepens, a capability that a competitor cannot simply replicate by signing up for the same subscription.

Sora is a useful reminder. So is Block. The tools and the roles that survive are the ones that build something that cannot easily reset. Choose your stack accordingly.

John McGann