Gold Dust
Two years in. Here’s what I’ve actually found — and the five questions that now govern every AI decision I make.
In 2023, we were told there was gold. In 2024, we were told we had the wrong shovel. In 2025, we were told we were digging in the wrong place. It’s 2026. I see gold dust. Not gold.
Gold dust is real. Not dismissing it. Employee productivity is up in specific pockets — data analysis runs faster, content gets a first draft in minutes, a few lines of software ship in hours instead of days. Actual, measurable, I-can-see-it value. But gold dust is not what we were promised, and it’s not what most companies are paying for.
Gold bars are what you take to the bank. Wealth-creating. Scalable. Structurally embedded in your P&L. That’s what two years of promises implied. Most of us haven’t found them.
Most CEOs answer the gap between dust and bar with more time, more adoption, more patience. Keep digging. Better shovels are coming. That answer has cost us something.
Global AI spending will hit $2.59 trillion in 2026 — up 47% from 2025, the fastest-growing technology expenditure in enterprise history. Enterprise GenAI alone went from $11.5 billion in 2024 to $37 billion in 2025. Still climbing. Somewhere in that number is Uber’s CTO, who burned through his company’s entire 2026 AI coding budget in four months.
Spending has doubled. Returns haven’t. Seventy-nine percent of companies report productivity gains. Five percent achieve substantial ROI. That gap — measurable improvement at the task level, almost nothing on the income statement — is what gold dust looks like from the boardroom.
BCG surveyed 1,000 CEOs in May. Sixty-one percent said their boards are rushing AI transformation. Half said job stability depends on delivering AI ROI. Thirty-five percent said AI outcomes are already baked into their performance reviews.
Pressure is baked into this gold rush. That’s not a neutral observation — it’s a force that will make you buy shovels you don’t need, on a timeline set by someone who isn’t digging.
I’ve been using AI for fifteen years. The startup I founded a decade ago ran on NLP — machine learning, natural language processing, well before large language models, before AI became something you had to have an opinion about at dinner. Not a newcomer.
So when 2023 arrived, not a skeptic. I felt the difference. LLMs were early, but the power was real — something genuinely new in the capability.
What I didn’t expect was the velocity of the vocabulary machine that followed. In three years, we’ve absorbed more new terms, frameworks, and paradigms than the prior decade combined. Each one delivered with urgency. Each one implying the last one was the wrong shovel.
Pitch has shifted every twelve months, without acknowledging the shift. 2023: deploy the chatbot, buy the Copilot. 2024: wrong tools — you need agents. 2025: the “year of the agent,” MCP, autonomous everything. OpenAI co-founder Greg Brockman called it: 2025 would be defined by AI shifting from chatbots to agents. Gartner’s response: forty percent of agentic AI projects will be cancelled by end of 2027. Now it’s 2026: large action models, AI stops observing and starts operating. New framing. Same messengers.
I didn’t restructure my team for AI. We made changes for business reasons. Other companies have used AI as the public explanation for decisions that look a lot like cost-cutting. I watched that. Made a different call. Fifty-five thousand jobs were cut industry-wide for AI in 2025 — twelve times more than two years prior. Forbes ran it on July 2: “AI Costs More Than the People It Replaced.” Fifty-five percent of employers who made those cuts already regret them. MIT found AI automation economically viable in only 23% of roles. Companies that cut the other 77% are running the math again.
The messengers have financial interests in where you dig. That’s not a conspiracy — it’s business. But it’s worth naming.
Finding the diagnostic question took time.
Get Baker Nanduru’s stories in your inbox
Join Medium for free to get updates from this writer.
“Is AI valuable?” isn’t it. Yes. Gold dust is real.
“Should we keep investing?” isn’t it. Depends entirely on what you’re investing in.
What cuts: what are you actually optimizing for?
Most companies optimize for the appearance of AI transformation — adoption metrics, AI features on the roadmap, the ability to say “we’re AI-forward” in a board presentation. Shovel-buying optimized for narrative, not outcome.
Companies landing gold bars — the 5% with substantial ROI, not just the 79% productivity bump — do something different. They dig where they found dust. Deeper. Rather than following the next narrative to a new location.
Gold dust is in three places for me: data analysis, content creation, software development throughput. Not everywhere — three places. That’s where the next dollar goes. Not a new shovel. Depth.
And here’s what makes the shovel math worse. My customers aren’t paying more for AI capabilities in the product. Two years ago, AI features were a differentiator. Now they’re table stakes — expected, not valued. I’m spending more on shovels to make the same money. The cost line moves. The revenue line doesn’t. If that ratio holds, this becomes a less profitable business with more durability risk — unless costs drop dramatically. Nobody’s modeling that in the pitch deck.
Five questions now run every AI investment I make:
1. What am I investing — and what am I sacrificing?
2. Where is my gold dust, and am I digging there?
3. How do I separate signal from narrative?
4. When do I invest more, hold, or stop?
5. Who is telling me the truth?
I’m not done digging. But I’ve changed what I’m digging for.
Gold dust is not a disappointment. It’s a signal — it tells you where the ground is fertile. Treating it as proof that the bar is just deeper, rather than as a map for where to look, is how you spend a third year buying shovels.
Every gold rush ends one of two ways. Either you find the bar — or you run out of places to tell miners to dig.
But there’s a third possibility I didn’t want to name until now. What if you find the bar — and it’s not gold anymore? Not because you were wrong. Because the world moved while you were digging. That’s the thought that keeps me up. The new reality has its own exchange rate, and nobody’s publishing it.
I’m still deciding which one this is. But I know this: I’m done buying shovels until I can answer all five questions.
