What Opus 5, MCP and Monday.com’s Layoffs Mean For Your Job

Anthropic cut Claude Opus 5’s price in half this week, the Model Context Protocol rebuilt its core plumbing for agents that scale, and Monday.com cut 600 jobs citing an AI-driven restructuring. Read together, these three stories say one thing: AI work is getting cheaper to build and easier to use as justification for cutting headcount.

Why did Opus 5 get cheaper?

Anthropic launched Claude Opus 5 on July 24, 2026 at half the price of its predecessor Fable, while claiming double the coding-benchmark score of Opus 4.8. That performance figure comes from Anthropic’s own launch announcement, and no independent lab had confirmed it as of this writing, so treat it as a vendor claim rather than a settled fact. The price cut is easier to verify: the same volume of tokens that cost you a given amount last month now costs roughly half that. For anyone who runs AI-assisted coding or research work daily, that changes what you can afford to automate without asking for a bigger budget. The bigger change sits underneath the price tag. Opus 5 ships with a new effort dial, low, medium or high, chosen per request instead of built into a single fixed model tier.

What does the effort dial change?

The effort dial in Claude Opus 5 lets you set low, medium or high reasoning depth on each request, according to Anthropic’s July 24, 2026 announcement. Low effort favors speed and cost on routine tasks like formatting or summarizing a document. High effort spends more computation working through a problem before answering, which Anthropic says produces the doubled coding-benchmark score over Opus 4.8. For someone who bills time or tokens against a project budget, this is the first frontier model where cost control sits inside a single request instead of forcing a choice between two separate products. The catch remains the same as above: Anthropic has not published third-party verification of the claimed gains, so run your own comparison before you trust the high-effort setting on work where a mistake would cost you.

Why is MCP going stateless?

The Model Context Protocol shipped its biggest specification update since launch on July 28, 2026, moving from a stateful session model to a stateless one. Stateful connections require an AI agent to keep a persistent session open with every tool server it talks to, which does not scale once a company runs dozens of agents against dozens of internal systems. A stateless design lets each server handle requests independently, closer to how ordinary web APIs already work, according to the protocol’s own blog post. If your team runs any MCP servers connecting agents to internal tools, this is not a cosmetic update. The old stateful handshake sits on a deprecation path, and integrations built against it will need review before that path closes.

What does that mean for your team?

If you build or manage AI agents connected to company tools through MCP, the stateless rework changes how those integrations authenticate and scale, per the July 28, 2026 specification update. Sessions that used to persist now need to be re-architected around individual stateless requests, which is engineering work, not a configuration tweak. Teams that treat this as optional maintenance risk finding their agent integrations broken once the old handshake is retired. The practical move is an inventory: list every MCP server your organization runs, flag which ones still assume a persistent session, and schedule the migration before a deprecation date forces it on you. No formal end-of-life date has been published yet, which makes this the week to start the list rather than the week to finish it.

Did AI cause Monday.com’s layoffs?

Monday.com cut 600 jobs and cited an AI-driven strategy change as the reason, according to TechCrunch’s running list of 2026 tech layoffs, published July 25, 2026. What that announcement does not prove is a causal link. A company citing AI as the reason for layoffs is also making a public relations choice, and TechCrunch’s list aggregates stated reasons from company announcements rather than independently verified causes. Monday.com joins a list of software companies restructuring around AI systems instead of only cutting costs, which is a pattern worth tracking across 2026. But treat the specific causal claim with caution: nothing in that report shows whether the eliminated roles were replaced by AI systems or simply reorganized elsewhere.

The three AI changes, ranked by urgency

1. MCP’s stateless rework hits first if you run agents. The July 28, 2026 specification update starts a deprecation clock on the old stateful handshake. If your job involves building or maintaining agent integrations, this is the item with a technical deadline attached, even before that deadline is formally announced.
2. Opus 5’s price cut hits second, on your next budget cycle. The July 24, 2026 launch makes frontier AI cheaper per task and adds a per-task cost dial. It affects what you can justify automating, but it will not force a decision on you this week the way a deprecation notice will.
3. Monday.com’s layoffs hit third, as a hiring-market data point. The 600 cuts reported by TechCrunch are one company’s framing of one decision. Useful to watch, not yet strong enough evidence to change your own career plan on its own.

What should you do this week?

Start with the item that has a clock running: if you or your team maintain MCP server integrations, write down which ones assume a persistent session before you do anything else. Next, if your work involves coding, drafting or research tasks priced by usage, run one real task through Claude Opus 5’s effort dial at both low and high settings and compare the output honestly rather than assuming the higher setting is always worth the cost. Last, read reports like TechCrunch’s layoff tracker as a prompt to ask better questions inside your own company, not as a verdict on your own role. None of these three stories tells you what to do with certainty. Together they tell you where to look first.

You are helping me evaluate whether a low-effort setting on Claude Opus 5 is good enough for a specific recurring task in my job. Here is the task I do repeatedly: [describe the task, e.g., drafting weekly status update emails, summarizing meeting notes, drafting boilerplate code]. Run through this task at low effort, then again at high effort if you're able, and tell me plainly where the two outputs differ: accuracy, completeness, tone, structure. Point to a specific line or decision where the two versions diverge rather than giving me a general verdict. If you need more detail about the task or my quality bar before you can do this well, ask me first.

You are a systems engineer helping me inventory AI agent integrations before a breaking protocol change. My team uses [describe which tools connect via MCP, or say you're not sure]. Help me build a checklist to find every MCP server we run, whether each one is stateful or stateless, and what would break if the older stateful session handshake were deprecated. For each server, tell me what information I would need to gather (session config, auth method, who owns it) and draft two or three questions I should ask the engineer who built it. Ask me clarifying questions about our stack before you produce the checklist.

Act as a careful analyst helping me interpret a tech layoff announcement without overreacting. Here is the situation: [paste the news item or describe the layoff, e.g., a company cut a number of jobs and cited an AI-driven strategy change]. Walk me through what this announcement does and does not prove: does it name which roles were cut, does it claim AI directly replaced people, is there any data beyond the company's own statement? Then tell me, based only on what's stated, what questions I would want answered before treating this as a broader trend in my own industry. Push back on me if I'm reading too much into a single data point.

Act as a blunt hiring manager who has read ten thousand resumes, not a career coach. I will paste my full resume and the job description I want. Rewrite the whole resume for that role, section by section, in this order: summary, experience, skills, education. Rules: every experience line leads with impact, not duty. Use bracketed placeholders like [8 percent] for any number I did not give you, and list at the end every placeholder I need to replace with a real figure. Keep it to one page of text. Plain formatting only, no tables or columns, so screening software can parse it. After the rewrite, tell me the three weakest claims that need evidence before I send this anywhere. My resume: [paste resume]. The role: [paste job description].