Council Post: Your Agents Are Reasoning In The Dark—Context Is The Light Switch
Jim Douglas is the Chief Executive Officer of Luciq, a leading agentic observability platform built exclusively for mobile.gettyMost organizations building agentic workflows aren't addressing the highest-lever...
Jim Douglas is the Chief Executive Officer of Luciq, a leading agentic observability platform built exclusively for mobile.
getty
Most organizations building agentic workflows aren't addressing the highest-leverage variable that impacts outcome and cost. They are investing in better models, better prompts and better orchestration frameworks while the impactful piece of the puzzle sits one layer below: the data the agent is operating on.
A pattern is emerging: The agent generates suboptimal results. The prompt gets rewritten. The evaluation continues. The actual problem never gets addressed.
The agent is only as good as what you feed it.
In an earlier piece, I wrote about how mobile teams are moving from reactive workflows to agentic ones. That shift is accelerating, but there is a prerequisite that does not get enough attention: the quality of the context those agents are operating on.
A general-purpose agent is a capable reasoner. What it cannot do is conjure information it was never given. If the problem lived at the device layer and the data was captured at the server layer, the agent is reasoning about a shadow of reality.
As explored in my company's article "The Queen Takes the King," most teams are feeding sophisticated agents shallow signals: fragmented telemetry that strips away the context needed to act. In mobile, where device-level signals are the difference between a fix and a guess, that gap is wider and more expensive than almost anywhere else in the stack.
Mobile is where the context gap hurts most.
Mobile is a fundamentally different operating environment. Device fragmentation, network variability, OS-level constraints, battery states and memory limits are not edge cases. They are the conditions under which your users experience your product every day. The data from our No Margin for Error report makes the stakes concrete: 83.4% of users rate stability as their top priority, and 15.4% will uninstall after a single crash.
Most observability tools were not built for this. They were built for infrastructure, surfacing server-side signals after the fact. That made sense when humans were reading dashboards. It does not make sense when agents are supposed to act autonomously. Feeding an agent an isolated stack trace produces two failure modes: Crucial edge-case signals stay invisible because no error was ever thrown, and agents starved of context hallucinate fixes based on assumed states rather than reality.
Context is not just data. It is structured, high-fidelity, agent-ready data.
McKinsey research published in 2026 found that "nearly two-thirds of enterprises worldwide have experimented with agents, but fewer than 10 percent have scaled them to deliver tangible value. Shaky data is often to blame; eight in ten companies cite data limitations as a roadblock to scaling agentic AI."
That is not a model problem. It is a data infrastructure problem, compounded further in mobile. The signals that explain what a user experienced do not live in the places most agent stacks are connected to.
Agents need context captured at the point of experience and mapped to the decisions they need to make. Every session, not a sample. Device state, interaction sequences, UI behavior, network conditions and sentiment signals, not just crash counts and error logs. Most organizations are trying to close this gap by throwing better prompts at worse data.
This organizational pattern is worth naming.
The teams struggling most moved fast on the agent layer without building the data layer first. The outputs were directionally plausible but operationally unreliable. Trust eroded, and the workflow never got redesigned.
The root cause is consistent: The data foundation was built to support human monitoring, not agentic action. There is also a compounding irony. Because coding agents let teams ship faster than ever, the volume of new code and simultaneous experiments has exploded. More code in production means more friction and a bigger ecosystem of signals to collect and track. Legacy monitoring tools collapse under that volume, and the productivity gain from shipping faster gets consumed by the maintenance burden of what was shipped. That is a reliable way to stall innovation entirely.
The leaders making progress are not asking, "Is our agent good enough?" They are asking, "Is the context we are giving our agent good enough?" The second question leads somewhere. The first mostly leads to another pilot.
The context layer is the most strategically important part of the agentic stack.
Models are commoditizing, and orchestration frameworks are multiplying. The data layer is not something you can spin up from a general-purpose integration. It takes time to build and domain expertise to get right.
A stack trace tells you where the app failed. It does not tell you why. Closing that gap requires 100% session capture across device conditions, interaction sequences, UI behavior and network context, because agents reasoning from incomplete data produce incomplete answers.
The teams getting this right early will have agents operating from a quality of context that general-purpose tools cannot replicate, and that advantage compounds.
The context gap is the real bottleneck in the agentic stack. Closing it is the work that matters most right now.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?