The Agentic Inflation Problem, or Do Agents have Agency?
Outcomes, not tools
There is a moment in the life of every powerful concept when its success becomes its undoing. The term achieves critical velocity — it appears in enough decks, enough announcements, enough panel discussions — and somewhere in that acceleration it loses its payload. It becomes a signifier detached from its signified. It becomes, in the precise sense, inflationary: more of it buys less meaning.
We are living through that moment with “agentic.”
Everything is agentic now. The chatbot with a search button. The workflow with an if/then branch. The assistant that sends an email on your behalf. The sophisticated multi-agent orchestration framework with memory, tool-use, and a feedback loop. All of it lands under the same banner, and the banner, overstretched, is beginning to tear.
I want to be precise about what I am not arguing. The agentic framing matters. It captures something real about the shift from AI-as-oracle to AI-as-actor — from systems that answer questions to systems that take actions, chain decisions, and operate across time horizons longer than a single prompt. That shift has genuine architectural, governance, and strategic implications. The vocabulary is not wrong; it has simply been emptied out by overuse.
What concerns me is what fills the vacuum left by a deflated concept: noise. And specifically, what gets crowded out.
Outcomes are getting crowded out.
The discourse around agentic AI is almost entirely architectural. We are absorbed in the infrastructure of agency — tools, context windows, memory layers, orchestration patterns, handoff protocols between agents. These are legitimate engineering and design problems. But they are means, not ends, and the conversation has drifted so far toward means that the ends have become almost invisible.
Further, Agentic thinking has become a superficial marketing gloss on delegation of responsibility. Agents have no agency of their own, and there is not much consideration as to the consequences of such delegation, especially when the outcomes are not as expected.
I rarely hear a crisp articulation of what agentic systems are actually changing. Not at the capability level — what they can do — but at the consequence level: what decisions are different, what costs are lower, what humans are freed to do something better with their time, what risks are introduced that weren’t there before. These are outcome questions, and they require a different discipline than the architectural ones.
Outcome-based thinking is not softer than agentic thinking. It is harder. It demands measurement where we have grown accustomed to demonstration. It requires someone to say, on the record: this process took forty hours and now takes four, and here is what that frees up, and here is what we’re watching. It forces a confrontation with the question of efficiency — not as a talking point, but as an obligation. Realtors and agents get paid only when they achieve the objective - the sale or purchase. We are still at the stage of staking out the property and setting the stage. This is only the start of the journey to realized outcomes.
The irony is that the tools now exist to actually deliver on this. The agentic moment is real. But without the discipline of outcome-thinking, we will build sophisticated machinery in service of poorly-specified ends, and then wonder why adoption is slower than the hype predicted.
The history of enterprise technology is littered with this pattern. “Digital transformation” was supposed to remake industries. Often what it remade was the org chart and the vendor roster. The transformation was architectural. The outcomes were assumed.
We should assume nothing this time. The question is not whether your stack is agentic. The question is what it’s for.
Pontifex covers strategy, technology, and the longer view. If someone forwarded this to you — subscribe to Pontifex here.
— Aaman Lamba
Writing from the bridge between worlds.
aamanlamba.com · Reply to this email — I read everything.
It’s all a set of standard moves in a cycle. We remember when there was the agile manifesto with good principles followed by everything becoming “agile”. The principles behind agentic are good, but it is now such a popular concept that there’s a tremendous market for people who don’t have the time/inclination to understand the principles but still want to adopt the shiny. As with all other things, the market will provide.
The term that comes to mind is “cargo culting”. Every technology trend, if it gets popular enough, gets cargo culted. There are plenty of corporate contexts where things may structurally locked in to the point that getting real value may not even be possible. That won’t stop them from adopting, and others helping them adopt.
It’s the early majority that are usually the worst offenders in cargo culting; as the late majority & laggards have usually waited long enough to miss the peak & fall.