Hey readers, bored yet?

Well the latest is in for Friday and big news… yayyy… another new release from Anthropic. Yeap, Claude Opus 4.8 released yesterday and the same week Anthropic confirmed a $30B annualized revenue run-rate, ahead of OpenAI's $25B.

KPMG, PwC, and EY each restructured their global delivery around AI agents in a three-week sprint that put more than 300,000 knowledge workers on Claude or Microsoft's stack.

Gartner published a survey of 350 executives that found 80% of companies running AI pilots cut workforce, with returns nearly identical between cutters and non-cutters. And at Sapphire, SAP put Claude inside Joule alongside 200 specialized agents.

The agent is becoming the unit of enterprise work, and the firms wiring it into how they sell, deliver, and price are pulling ahead of the ones still treating it as a pilot.

By the end of the newsletter you’ll see also a free resource of claude skills and plugins to be used to run a business in Kosovo or outside of Kosovo, using Claude Code, built by 38shift.com

Anthropic released Claude Opus 4.8 on Thursday, the same week it confirmed a $30B annualized revenue run-rate ahead of OpenAI's $25B. The model posts 69.2% on SWE-Bench Pro (up from 64.3% on 4.7), 88.6% on SWE-bench Verified, and 83.4% on OSWorld-Verified for computer use.

It went live day-one on AWS, Google Cloud, and Microsoft Foundry, and Anthropic teased Mythos-grade alignment for the broader release window.

Three things change in practice. Effort Control on claude.ai gives you a dial from default to "extra" to "max," so harder tasks get more compute without switching SKU. Dynamic Workflows inside Claude Code spawns hundreds of parallel subagents in a single session and runs codebase-scale migrations across hundreds of thousands of lines.

Fast Mode for Opus 4.8 runs at 2.5x speed at one-third the previous price. The qualitative shift Anthropic is selling: 4.8 is roughly 4x less likely than 4.7 to leave flaws in its own code unflagged.

If you ship software, the question is no longer "can a model write this function." It is "how do I structure my repo so 200 parallel subagents can chew through it overnight."

Claude Code is already a $2.5B run-rate business with 54% of the enterprise coding market, and Opus 4.8 widens the gap between teams that have rewired their development around agents and teams still treating Claude as a faster autocomplete.

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

KPMG and Anthropic signed a global alliance on May 19 embedding Claude inside Digital Gateway, the Azure platform 276,000 KPMG staff use for client work. Five days earlier, PwC committed to training 30,000 staff on Claude Code and Cowork.

Two days after KPMG, EY and Microsoft announced a $1B, five-year alliance to embed EY consultants alongside Microsoft engineers inside a 15-country AI Factory. Three of the Big Four rewired delivery around agents in one three-week window.

The Big Four are the canary for every knowledge-work firm of any size. The audit class moves first because clients are pricing the work differently and the firms are matching it before margins erode.

These are not chat-tool seats. They are integrations at the work-platform layer, with PwC reporting insurance underwriting cycles compressed from ten weeks to ten days, and KPMG saying agent build time for tax workflows dropped from weeks to minutes. The unit being repriced is the engagement.

Three facts changed at once for anyone running a services business or a knowledge-work team. Your buyers now have a benchmark for how fast their advisors can deliver. Your hires have a benchmark for what tools they expect on day one. Your competitors who are not announcing similar deployments will look slow inside the next two RFP cycles.

The question is not whether you sign with Anthropic or Microsoft. It is what you ship this quarter to match the cadence the Big Four just set.

Gartner surveyed 350 executives at $1B+ companies that had piloted or deployed AI agents and autonomous tech. About 80% reported workforce reductions tied to the work. The headline finding: those reductions did not translate into ROI. Cutters and non-cutters reported nearly identical returns. Helen Poitevin's line from the press release: "Layoffs may create budget room, but they do not create return."

This is the contrarian fact pattern of the year. The companies winning are the ones using AI as "people amplification", adding lift to existing workers rather than substituting for them.

The 4x improvement on unflagged bugs in Opus 4.8 is the example: it does not make engineers redundant, it makes one good engineer review the output of a hundred parallel subagents instead of writing it themselves. The return shows up where the human is still in the loop.

Sit with this the next time a "cutting to invest in AI" memo crosses your feed. Some of those cuts are real productivity rebalancing.

Most are capital reallocation dressed in narrative, and Gartner is saying the dressing comes off in the numbers.

The teams compounding right now are using AI to ship 10x with the same headcount, not 1x with one-tenth of it.

The headcount line on a slide is not the same as a return on the investment that funded it.

At Sapphire on Wednesday, SAP launched the Autonomous Enterprise: 50 Joule Assistants coordinating more than 200 specialized agents across finance, supply chain, procurement, HR, and CX.

Claude sits as a named foundation model inside the stack, next to Mistral, Cohere, and Google's Gemini.

The user interface is Joule Work, where employees describe outcomes and Joule routes the work across systems.

Two days later, LAUNCHub Ventures hired Vedran Blagus out of South Central Ventures to lead Croatia, Slovenia, and the Western Balkans.

The two moves are the same trade. SAP is admitting that the unit of enterprise software is no longer the module, it is the agent, and the agents ride on frontier models the company does not own.

LAUNCHub is admitting that the next decade of CEE deals lives in Zagreb, Ljubljana, Pristina, and Belgrade as much as in Sofia. Both decisions are about positioning closer to where the value gets created, before it does.

For founders in the region, the SAP-Anthropic alignment is the bigger gift. Every customer SAP signs is a customer who will need someone to build the Joule agents that actually work for their business. That layer is open, it is local, and it is the wedge a small CEE team can take without competing with a hyperscaler. The Blagus hire is the capital telling you it sees the same thing.

Productivity: Effort Control on Opus 4.8 gives you a real throttle. Claude.ai added a dial today: default, extra, max. Use max on memos, deep analysis, contract review, and any prompt where you want it to keep working past the first answer. Default keeps casual back-and-forth snappy. Treat it like throttle, not a once-and-forget setting. The price is the same; the patience is yours to allocate.

Design: Anthropic MCP tunnels lets your design library stay inside the firewall. Anthropic's MCP tunnels research preview (shipped May 19) lets Claude reach internal Figma libraries, brand tokens, and component systems without your assets leaving the security perimeter. If legal has been blocking AI design work because of asset egress, this is the unblock to bring to the table this week.

Dev: Claude Code Dynamic Workflows is the first parallel agent worth pointing at a real repo. Open Claude Code, point Opus 4.8 at a monorepo migration you have been postponing, and ask for a plan. It will spawn the subagents it needs and report back. The first real test is a dependency upgrade across services; the second is a typed-language rewrite of a hot path you never had the time to touch. Let it run overnight.

Dev: Camunda ProcessOS turns your existing workflows into agentic ones. Camunda ProcessOS opened its closed beta on May 20. It points at your existing BPMN processes, finds the steps an agent should own, and rewrites them in place. If your team has been waiting for a way to add agents without ripping out the orchestration layer, this is the lowest-friction entry point shipped this quarter.

A Cowork plugin marketplace and landing page that helps Kosovo small businesses run their back office. The original plugin serves freelancers and IT consultancies (pitches, contracts, invoices, taxes); three vertical plugins extend the same Kosovo-grounded approach to law firms, restaurants, and financial firms.

Check out the project here dixhitale.com

Or the repo: https://github.com/celiknimani/biznesi-dixhital