I'm Not in Tech. I'm in Mining. And Everything Matt Shumer Wrote Is Already My Reality.

Originally published on LinkedIn on 18 February 2026.

Matt Shumer's essay "Something Big Is Happening" went viral this week — over 100 million views and counting. His core thesis: we're in the "this seems overblown" phase of something much bigger than Covid, and the people closest to AI are trying to warn everyone else before it's too late.

I read it and felt one thing: recognition.

Not because I'm an AI founder or Silicon Valley investor. My nominal day job is Principal of Continuous Improvement at a mid-tier gold mining company: I work on management operating systems, operational excellence, and continuous improvement at mine sites across the US, New Zealand, and the Philippines.

If there's an industry that's supposed to be the last to feel this wave, it's mining.

And yet, everything Shumer describes — the moment AI went from "useful tool" to "this changes everything" — already happened to me. Not in theory. In practice. Over the past year and a half, I've gone from asking Claude the occasional question to building an AI infrastructure that has fundamentally changed how I work, how I build, and what I'm capable of producing.

I want to share what that actually looks like — in granular detail — because I think Shumer's message is more urgent than most people realise. And the most compelling evidence isn't in Silicon Valley. It's in the most traditional industry on earth.

Some Context: I'm Technical, But I'm Not a Developer

I want to be honest about my starting point, because it matters. I've always been tech-inclined. I used to code a fair bit in Python and C#. I built automations, dabbled in data pipelines, and was generally the person in the room who understood what was possible with technology. But I didn't make my career out of software development. I made it out of continuous improvement — process optimisation, management systems, operating excellence. My career has been spent in mining operations, not in IDEs.

"It's not about developers being replaced. It's about everyone gaining the ability to build."

What AI did was take that predisposition and curiosity and turn it into something completely different. A year ago, I could write scripts. Today, I'm building full agentic systems — multi-agent orchestration, autonomous trading algorithms, enterprise AI strategy — at a pace and quality level that would have required a dedicated development team eighteen months ago.

The gap between "tech-curious domain expert" and "building production systems" has collapsed. That's the part of Shumer's essay that hit hardest. It's not about developers being replaced. It's about everyone gaining the ability to build.

The Infrastructure: My Personal AI Operating System

It started with Obsidian — a plain-text note-taking tool — and a curiosity about whether I could connect it to Claude. That curiosity turned into a system that now touches nearly every part of how I work and live.

The connective tissue is MCP — the Model Context Protocol — a standard that lets AI models connect directly to external tools and data sources through purpose-built servers. Instead of copy-pasting into a chat window, the AI reaches into your systems, reads your data, and takes actions on your behalf.

I built custom MCP servers — hosted on my own infrastructure, routed through Cloudflare, networked via Tailscale — that give Claude direct access to:

• Obsidian vault (Knowledge Base) — seven years of daily journals, project notes, frameworks, and ideas. Claude can search, read, create, and update notes across the entire vault.
• Spending patterns (UP Bank) — transaction history, spending categories, account balances. Access to actual financial data for analysis.
• Health data (Apple Health) — sleep patterns, weight trends, activity levels, mood tracking. Claude correlates sleep quality with journal entries about stress periods.
• Meal planner (Mealie) — recipes, weekly meal plans, macro targets, shopping lists. Claude generates meal plans respecting family rules about prep time and protein rotation.
• Email, calendar, and contacts (Google Workspace) — consolidated into a single server. Claude checks schedule, searches inbox, drafts responses, and coordinates.
• Code repositories (GitHub) — Claude searches repos, reads code, creates issues, and reviews pull requests across all projects.
• Task manager (Todoist) — full integration for creating, updating, and querying tasks across personal and work projects.

All of this runs on infrastructure built and maintained independently. A year ago, I wouldn't have known how to build any of it. The AI helped build the infrastructure that AI agents now use to develop and extend other AI agents.

That's Shumer's recursive loop, playing out in a mining professional's home office in suburban Brisbane.

The Work System: Agent Delegation From Issue to Completion

The personal infrastructure was the training ground. What came next was the professional application — and this is where things get genuinely transformative.

"I work now with a hybrid team of people and AI agents."

I manage projects through Linear — a modern project management tool — with AI coding agents assigned as collaborators on issues. Not as a gimmick. As actual team members who receive work, execute it, and deliver completed pull requests for review.

The workflow: I create a well-scoped issue in Linear with clear acceptance criteria. I assign it to an AI agent — FrontEnd Backend Coding Agents, Design Agents, Testing Agents — all in the ecosystem. The agent spins up a sandboxed environment, clones the work required, implements the solution, runs tests, and opens a pull request. Agents review the PR, provide feedback if needed, and merge. Tasks that would have taken a weekend of coding are completed while sleeping, with time allocated only for focused direction work.

I have repositories for:

• An autonomous cryptocurrency trading system
• A mining M&A due diligence agent built on Google's Agent Development Kit
• Multi-agent orchestration experiments coordinating multiple AI coding agents in parallel

This Linear workspace has dedicated projects for each, with issues flowing from backlog to completion through a pipeline where AI agents build and direction is provided. This happens at scale, across multiple projects, every day.

This isn't speculative. Linear issues assigned to different agents are marked "completed" with merged pull requests attached. The agent reads the issue description, understands codebase context through custom instructions, and delivers working code. When corrections are needed, reviewing agents or myself provide feedback on the PR and it iterates — exactly as with a junior developer.

Code is only one deliverable category. The same delegation model applies to everything produced: presentations, analyses, technical papers, data visualisations — all flow through Linear with well-scoped issues and clear acceptance criteria. The difference is whether the agent is a coding agent opening a PR, or Claude working through Cowork to produce formatted PowerPoint or Excel analysis. The pattern is identical: define deliverable, assign work, review output.

What a Typical Day Actually Looks Like

4:30 AM — Plan the day. I don't open a blank journal. My morning briefing is already prepared. It reads my Obsidian planning cascade (annual goals → monthly targets → weekly intent), pulls today's calendar, checks Todoist for overdue and priority tasks, reviews last night's sleep and health data from Apple Health, and synthesises all into a morning briefing. The output is a grounded, personalised plan — informed by real schedule, real sleep score, real project status, and actual quarterly goals. It writes directly into my daily journal.

During work — Triage and delegate. The working day functions as a triage function with a roster of custom agents. A Work Planning Agent automatically runs across all Linear boards and Todoist lists, scopes what needs to happen, and assigns work. A presentation needs updating? Create an issue with source documents, acceptance criteria, and brand conventions, then delegate to Claude via Cowork. The agent searches the Obsidian vault for relevant project notes, pulls data from drawings, storylining, and meetings, references the Linear project board for current status, and produces a formatted draft. A coding task needs implementation? Assign to GitHub Copilot or Jules. An analysis needs building? Same pattern — issue, scope, delegate. The role becomes direction and quality control, not execution.

Coding blocks — On weekends, there's a protected four-hour window. In that time, scope issues in Linear, assign them to AI agents, and ship features that would have taken a development team a sprint to deliver. Review PRs, provide direction, and merge. The agents handle implementation.

Meal planning — On Sundays, Claude generates a weekly meal plan through Mealie that hits macro targets for both participants from a catalogue respecting prep time rules and protein rotation, and generates a shopping list organised by store section.

Evening — Finish the day. Sisyphus (an agent) cross-references GitHub activity (commits pushed, issues closed, PRs merged) with Todoist completions, filters out routine items, and produces a narrative summary of meaningful progress. It writes directly into the journal alongside the morning briefing, closing the loop on the day. Review it, reflect briefly, and wind down.

"This isn't a demo. This is Tuesday."

The Actual Productivity Shift

The actual work of my job takes minutes now, not hours.

Not because the job is simple. Because the leverage is extraordinary. Analysis that used to require manually gathering data from multiple sources, cross-referencing in spreadsheets, and producing formatted reports now happens through a system where Claude accesses data sources directly, performs analysis, and produces output. The role has shifted from doing the work to directing it — scoping what needs to happen, reviewing what was produced, and making judgment calls about quality and direction.

I am more productive than ever in my life. And I'm not working harder. I'm working differently.

"The old ways of using Excel, Power BI, SQL, PowerPoint manually are dead and they are never coming back for me."

What's happening in the personal workflow is a preview of what's coming for organisations.

The killer argument for enterprise AI adoption isn't about benchmarks. It's about capability compounding. Each tool built — each plugin, each MCP server, each agent workflow — becomes a building block for the next one. The organisation that starts now doesn't just save time on today's tasks. It builds an infrastructure that makes tomorrow's tasks trivially easy.

Mining companies are spending hundreds of millions on operational improvements. An AI capability that reduces analytical cycle time by 80% and enables a single person to do the work of a small team isn't a nice-to-have. It's a competitive moat.

The Gap Between Perception and Reality Is Dangerous

The thing that resonates most about Shumer's essay is his frustration with the "cocktail party version." That feeling is familiar.

When colleagues in mining hear about this work with AI, they typically respond with either polite scepticism — "that's interesting, but our industry is different" — or genuine curiosity followed by "I wouldn't even know where to start."

Both reactions miss the point. The industry isn't different. The timeline is. Mining will feel this wave later than software engineering, but the water is already rising. And "I wouldn't know where to start" is exactly the problem Shumer is trying to solve. You start by subscribing to the best model available, and you start pushing your actual work into it. Not hypothetical prompts. Your real documents, your real data, your real decisions.

I went from zero MCP servers to a full personal AI infrastructure in about eight months. I went from "Claude, explain this concept" to "Claude, check calendar, review project board, search vault for notes, draft paper, and file summary when done." The gap between those two states is enormous. And it's accelerating.

This Is Not About Replacing People

I'm not being replaced. I'm being amplified.

My domain expertise — understanding mining operations, continuous improvement methodologies, how organisations actually change — is more valuable now, not less. What's changed is the leverage. Ideas that used to stay in my head because I lacked technical skills to build them are now real systems. Analysis that used to take weeks takes hours. Communication that used to be a bottleneck is now assisted by an AI that knows context, style, and objectives.

The people who will struggle aren't the ones whose jobs get "automated away." They're the ones who refuse to pick up the tool. Because the gap between an individual who uses AI effectively and one who doesn't is already enormous — and it's compounding daily.

What I'd Tell My Colleagues

If you're reading this from a traditional industry — mining, energy, agriculture, manufacturing, construction — here's what I'd say:

You are not too late. Most of your peers haven't started. The window of advantage is still wide open.

Start with the paid tier. Free ChatGPT is a year behind what paying users have access to. Subscribe to Claude or ChatGPT Pro. Use the best model available — right now that's Claude Opus 4.6 or GPT-5.2, but it changes every few months.

Feed it your actual work. Not "write me a poem." Give it your quarterly report. Your technical document. Your project plan. Your messy spreadsheet. See what happens when you do something real.

Build the muscle. The first week will feel clunky. By the second month, you won't be able to imagine working without it. By six months, you'll be building things you didn't think were possible.

Think in systems, not prompts. The real power isn't in asking better questions. It's in connecting AI to your data, your tools, and your workflows. That's where the compounding starts.

Shumer is right. Something big is happening. I know because it already happened to me — not in Silicon Valley: but in Brisbane where a mining professional built an AI infrastructure that would have been science fiction two years ago.

The water is rising. You can feel it if you're paying attention. Start swimming.