Either the AI bubble is about to burst... or the economics of knowledge work are being rewritten.
I don't know which one we're in. And I'm increasingly skeptical of anyone who claims they do.
The Numbers Say One Thing
Forrester published its latest AI Job Impact Forecast in January 2026, predicting that AI and automation will displace about 6.1% of US jobs by 2030 — roughly 10.4 million positions. That's significant (comparable in scale to the Great Recession), but it's not the extinction-level event you'd expect from the headlines. The report also makes a point worth sitting with: most of the layoffs being attributed to AI right now are financially driven. AI is the scapegoat, not the cause. Nine out of ten times, when a CEO says "we're replacing headcount with AI," there isn't even a mature AI application ready to fill the gap.
So by that measure, maybe this whole thing is being overstated.
Source: Forrester — AI And Automation Will Take 6% Of US Jobs By 2030
The Shipping Velocity Says Something Else
Then I look at what's actually shipping — not what's being promised, but what's landing in production — and the picture changes.
OpenAI's GPT-5.x Codex line is functioning less like a model you prompt and more like an autonomous coding surface. These agents can reason across entire repositories, iterate on their own code, and run for hours unsupervised. By the end of 2025, Codex had moved from "a model you ask for help" to a coding teammate that connects models, local tooling, and cloud infrastructure to tackle long-horizon, project-scale work. Anthropic delivered Claude Co-Work — a desktop tool for non-developers to automate file and task management — in roughly 10 days. OpenAI itself noted that it releases a new feature or capability approximately every three days.
That gap between "6% of jobs by 2030" and "new capabilities every 72 hours" is what's genuinely hard to reconcile.
The Discourse Is Loud. And Mostly Unhelpful.
Matt Shumer's essay "Something Big Is Happening" has crossed 82 million views. He compares this moment to pre-COVID — where the signals were there but most people didn't internalize them until it was too late to prepare.
Link: Matt Shumer — Something Big Is Happening
The skeptical rebuttals are out there too, and they make fair points about hype cycles, overextrapolation, and the gap between demos and deployed value. MIT researchers have noted that 95% of generative AI pilots at companies are failing to positively impact P&Ls.
Link: Mashable — Skeptical Take on "Something Big"
Both sides, if I'm being honest, sound a little extreme. The truth — as usual — is somewhere in the messy middle.
What I'm Actually Seeing in the Work
Meanwhile, in the real world — in actual day-to-day work — things are shifting in a very practical way.
I run a small design and engineering firm. We build solutions for federal agencies and commercial enterprises. And over the past several months, I've watched our productivity increase by an order of magnitude — not through some grand AI strategy, but through steady, deliberate adoption of tools across every phase of how we work.
Development cycles are shorter. Refactoring is faster. Writing test cases isn't the same task it used to be. Documentation doesn't take half a day anymore. Marketing drafts don't either. Code reviews that used to take a morning now take an hour. Prototypes that took a week take a day. Proposal sections that used to require days of research and writing now have solid first drafts in hours.
But it's not just the engineering side. We're using AI to streamline business operations — contracts, compliance documentation, competitive analysis, internal knowledge management. The work that used to eat up a small team's bandwidth is getting compressed in ways I wouldn't have believed two years ago.
This isn't theory. It's measurable time compression. And it's happening across roles, not just engineering.
The Part That's Harder to Ignore
What's really worth paying attention to is where the community is heading — because the trajectory tells you more than any forecast.
Ralph (ghuntley.com/ralph) is a technique — essentially a bash loop that keeps an AI coding agent running continuously, iterating on its own output, testing, failing, tuning, and shipping. In its purest form, it's while :; do cat PROMPT.md | claude-code ; done. One engineer reportedly delivered a $50K contract scope for under $300 in compute. Ralph is currently building an entire programming language — one that wasn't in any LLM's training data — and programming in it. It's deterministic in its imperfections, and those imperfections are tunable.
OpenClaw (openclaw.ai) is an open-source personal AI assistant that sits on your own machine and operates through WhatsApp, Telegram, Discord, or Slack. It manages your email, calendar, files, browser, and code — 24/7 — like a digital employee you message as a coworker. It has persistent memory, full system access, and the ability to write its own plugins. It's less than a month old and already has people running it as a company-level assistant, not just a personal tool.
These aren't enterprise platforms yet. They still have a road to travel before they're secure, governed, production-grade solutions. But the trajectory is unmistakable: AI is moving from "tool you prompt" to "agent that works for you autonomously" — doing knowledge work for hours, days, even weeks, as a proxy for a human. The concept of a "digital employee" that can operate independently, manage its own context, and deliver completed work isn't science fiction anymore. It's a GitHub repo.
Where I Actually Land
So when people say "bubble," I don't fully buy it. The underlying capability is real and it's compounding.
When people say "mass extinction of jobs," I don't buy that either. Forrester's own data shows AI strongly influencing jobs (20%) far more commonly than replacing them (6%) — and even that replacement number assumes organizational readiness that most companies don't have.
What I see is something more fundamental: a shift in what it means to do knowledge work.
We've seen this before. When mechanized looms arrived in the early 19th century, skilled weavers didn't vanish — but the nature of their work changed completely. The craftsman who once controlled every thread became the operator who managed machines that did the weaving. The work didn't disappear. The relationship to the work did. New roles emerged that no one had language for yet: machine operators, factory supervisors, maintenance engineers, industrial designers. The overall economy grew. Productivity climbed. And the people who adapted earliest had the most agency over how their careers evolved.
We're in a similar moment. AI tools are not eliminating knowledge work — they're restructuring the relationship between the worker and the work. The shift isn't from "employed" to "unemployed." It's from "person doing the work" to "person getting work done with AI." At our firm, the work itself hasn't gone away — the contracts, the deliverables, the client requirements, the quality standards. What's changed is how we get it done. The process has been rewritten while the outcomes remain the same (or improve).
The Skills That Matter Next
And just like the industrial revolution, this transition is creating an entirely new skills landscape.
McKinsey just gave us a live case study. The firm now has 25,000 AI agents working alongside its 40,000 human employees — added in under two years. CEO Bob Sternfels described their model as "25-squared": grow client-facing roles by 25%, shrink back-office roles by the same amount. The result? Non-client-facing headcount went down, but output from that group went up 10%. The firm saved 1.5 million hours of work last year. That's not a theory about the future — that's a top-tier consultancy restructuring the relationship between people and work in real time.
Source: Business Insider — McKinsey's 25,000 AI Agents
The broader labor market data tells the same story. Job postings requiring AI fluency have grown sevenfold in just two years, according to McKinsey — the fastest-growing skill category in U.S. job postings. Over half of those postings were outside IT and computer science, spanning marketing, finance, HR, and operations. PwC found that skills in AI-exposed occupations are changing 25% faster than in less exposed roles — and carry up to a 25% wage premium. The market isn't just asking for AI specialists. It's asking every knowledge worker to become AI-literate.
Here's the part that doesn't get enough attention: there are real limitations to what AI can do today. It hallucinates. It lacks judgment. It doesn't understand your business the way a ten-year veteran does. It can't navigate organizational politics, read a room, or make the call when the data is ambiguous. Those constraints are exactly why the shift is toward working with AI rather than being replaced by it. The most effective pattern we've found isn't "hand it to the AI" — it's human direction with AI execution. The person still sets the strategy, catches the errors, and owns the outcome. The AI accelerates the getting-there.
Will there be churn? Absolutely.
Some roles will compress or disappear, the same way the typing pool disappeared after word processors. But new roles are already emerging — AI engineers, prompt specialists, human-AI interaction designers — roles that didn't have names two years ago. Autodesk's 2025 AI Jobs Report found that AI Engineer positions grew 143% and Prompt Engineer grew 136% year over year. And notably, design skills have overtaken pure technical expertise as the most in-demand skill in AI-related job postings — a signal that creativity, communication, and judgment are becoming more valuable, not less.
The industrial revolution didn't make human labor irrelevant. It changed what humans needed to be good at. The AI revolution is doing the same thing — faster, and across a wider surface area of the economy. The question isn't whether your role will be affected. It's whether you'll be the one who shapes how it evolves, or the one who gets reshaped by it.
What This Means for Government Contracting
I work in this space, so I'll say what I think a lot of people in the GovCon world are sensing but haven't fully articulated yet: AI is going to reshape this sector in ways we can't fully predict — and the disruption won't be evenly distributed.
Some agencies will move fast. The ones with modernization mandates, flexible contract vehicles, and leadership willing to experiment are already pulling AI into mission-critical workflows. Others will lag — constrained by legacy systems, rigid contract structures, and compliance frameworks that weren't designed for a world where an AI agent can do in hours what a team used to do in weeks. That gap between early adopters and laggards within the federal landscape is going to widen before it narrows.
But the more fundamental shift is in what the government will buy — and from whom. As agencies get smarter about what AI can deliver, the nature of the services they seek will evolve. The demand for staff augmentation — putting bodies in seats — will face pressure as AI-assisted teams demonstrate they can deliver equivalent or better outcomes with fewer people. Contracts that once valued headcount will start valuing velocity, quality, and outcomes. That changes the math for every contractor in the ecosystem.
On the contractor side, the firms that adapt will look different. The services they offer, the way they structure teams, and the skill profiles they hire for are all in motion. Firms that figure out how to blend human expertise with AI capability — securely, within FedRAMP and compliance constraints — will have a meaningful edge. Those that don't will find themselves competing on a playing field where their traditional advantages matter less every quarter.
There's a much deeper analysis to be done here — the intersection of AI, federal acquisition, compliance, and workforce transformation deserves its own treatment. I'm planning to write that piece. For now, the signal is clear: the GovCon sector isn't exempt from this shift. It's just moving on its own timeline.
The Only Prudent Move
I don't have a grand thesis here. I'm not selling anything. I'm not predicting doom or utopia.
But I will say this: you cannot learn to swim by reading books about swimming. At some point, you have to get in the water.
That's the best analogy I've heard for where we are with AI right now. You can read every forecast, follow every debate, bookmark every essay — and still have no idea what these tools actually mean for your work until you use them. The gap between understanding AI conceptually and understanding it through practice is enormous. The patterns only reveal themselves when you're in the middle of it — when you see an agent save you three hours on something you thought would take all day, or when you realize a workflow you've done manually for years can be compressed into minutes. Those aren't insights you get from the sidelines.
So the prudent move isn't just to "keep up." It's to actively engage. Pick a tool. Apply it to a real workflow — not a demo, not a sandbox, the actual work. Measure what changes. Learn the patterns. Understand where it's strong and where it falls apart. Build your own informed perspective instead of borrowing someone else's.
The people who will navigate this transition well — whether it's a bubble, a leap, or something in between — won't be the ones who predicted it correctly. They'll be the ones who learned by doing, adapted as the landscape shifted, and made rational decisions grounded in their own experience rather than someone else's headline.
Curious what's actually changing in your day-to-day — not the headlines, the work.
The perspective, experiences, and editorial judgment in this article are entirely mine. Research, drafting, and iterative refinement were done collaboratively with AI. I directed the thesis, shaped every section, and made all editorial calls — the AI helped me get there faster. Which, if you think about it, is kind of the point of the article.
Originally published on LinkedIn.