# Why AI Won’t Fix the Hardest Part of Building Connected Products
## Introduction
The rise of AI-assisted development has fundamentally altered what product teams can accomplish in a sprint. Interfaces that once took weeks to sketch out can now materialise in hours. Prototypes that required months of coordination are standing up before lunch. For software teams everywhere, this acceleration feels like a genuine inflection point.
But in the world of connected products—smart devices, IoT ecosystems, systems that bridge multiple networks and data sources—there is a quiet realisation spreading through engineering and product leadership. The software itself was never the bottleneck. The real difficulty lives in the spaces between systems, in the reliability of data, and in the trust that customers place in a product that has to orchestrate dozens of invisible handoffs just to work.
This article explores why AI productivity gains, while real, do not eliminate the hard problems that define connected product development.
—
## The Illusion of Simplicity
There is a persistent misconception that connected products are essentially software projects with a hardware layer attached. The reality is far more nuanced. A connected product is, at its core, a web of agreements between disparate systems that must communicate seamlessly under unpredictable real-world conditions.
Consider a smart energy platform that connects household solar panels, battery storage, grid pricing signals, and consumer behaviour. The code that drives each of these components may be sophisticated but ultimately straightforward. The difficulty lies in reconciling inconsistent data formats, handling outages gracefully, and ensuring that when a user sees a “fully charged” indicator, the system has actually completed the task reliably.
This kind of operational complexity cannot be generated or refactored away by an AI tool. It must be understood, designed for, and continuously maintained through careful engineering judgment.
—
## Standards Are Necessary but Not Sufficient
Industries that deal with connected ecosystems often adopt common protocols to encourage interoperability. These standards provide a shared vocabulary that different vendors can implement. However, the gap between having a standard and achieving dependable, real-world compatibility is vast.
Two systems can both claim compliance with the same protocol while interpreting its requirements differently. One vendor may implement a particular message format strictly; another may leave room for ambiguity. When these systems are asked to communicate at scale, the inconsistencies surface not as obvious errors but as subtle, intermittent failures that are extraordinarily difficult to diagnose.
A product team building on top of these standards inherits the burden of making disparate implementations behave as though they were a single coherent system. This requires deep domain knowledge, extensive testing across edge cases, and a willingness to build compensating logic where the standard falls short. No amount of AI-generated code removes this responsibility.
—
## Data Quality Is the Foundation Everything Rests On
In any connected product, the decisions the system makes are only as good as the information it receives. Bad location data misleads a logistics platform. Incomplete sensor readings distort a predictive maintenance model. Stale inventory signals cause a supply chain tool to recommend actions that no longer reflect reality.
What makes data quality especially challenging in connected systems is that it is not a single-point problem. Data flows through multiple stages—collection, transmission, transformation, and presentation—and degradation can happen at any point along that chain. A sensor may report accurately, the network may introduce latency, and the processing layer may misinterpret a timestamp. The end user sees a wrong result without ever knowing where the chain broke.
AI tools can help process and transform data more quickly, but they cannot distinguish a trustworthy source from an unreliable one. That judgment requires human expertise and a clear understanding of where the data originates and what biases or gaps it carries.
—
## When Speed Becomes a Trap
One of the more subtle risks of AI-assisted development is that it lowers the cost of experimentation so dramatically that teams can be tempted to move fast without going deep. It becomes tempting to build feature after feature, ship quickly, and iterate later. But in connected products, a poorly considered addition can introduce instability across an entire ecosystem.
A new integration point, a change in how data is processed, or even a minor tweak to a user-facing workflow can ripple outward in ways that are difficult to predict. Without a solid understanding of the customer’s actual needs and the operational environment where the product lives, speed becomes a liability rather than an advantage.
The most effective product teams use the time saved by AI tooling to do the harder, less glamorous work: observing users in their real environment, stress-testing assumptions about how a product will perform under adverse conditions, and prioritising robustness over novelty.
—
## The Real Bottleneck: Understanding Before Building
If the code itself is no longer the constraint, what is? The answer is insight. Knowing which problems deserve solving, understanding the context in which users interact with a product, and designing experiences that account for the way people actually behave rather than the way they theoretically should.
A connected home thermostat, for example, is not really a thermostat. It is a system that must learn household routines, respond to external weather conditions, communicate with utility pricing signals, and present a simple interface to someone who has no interest in any of that complexity. The product’s success depends entirely on how well the team understands the human at the centre of the system.
This kind of understanding comes from time spent with users, not from faster code generation. It comes from watching someone struggle with a setup process, from noticing that a notification arrives at the wrong moment, from hearing that a feature works technically but feels unintuitive. AI can help express these insights in code once they are found, but it cannot find them in the first place.
—
## FAQ
### Does AI help at all with connected product development?
Absolutely. AI tools excel at accelerating routine development tasks, generating boilerplate code, assisting with testing, and helping teams prototype quickly. The benefit is real and significant. The key is to recognise where AI adds value and where it cannot substitute for deep domain understanding, data integrity work, and product judgment.
### Are standards like MQTT, OCPI, or HTTP enough to guarantee interoperability?
Standards provide an essential starting point, but they do not guarantee smooth operation at scale. Different implementations, partial compliance, ambiguous specifications, and edge-case behaviours all create gaps that product teams must bridge. Standards reduce the complexity of building integrations but do not eliminate the need for careful validation and testing.
### How do product teams identify which data sources to trust?
Trusting data sources requires a combination of domain expertise, historical analysis, and ongoing monitoring. Teams should establish clear data lineage, validate sources against known ground truth where possible, and build automated alerts when data patterns deviate from expectations. This is fundamentally a human-driven process supported by tooling, not a task that AI can solve independently.
### What should a product team prioritise when using AI tooling for connected products?
The highest-impact use of saved time is investing in customer discovery, real-world testing, and reliability engineering. Building faster is only valuable if what is being built addresses genuine needs and works reliably in production conditions. Teams should resist the temptation to equate velocity with progress and instead focus on outcomes that matter to users.
### Can AI improve data quality in connected systems?
AI can assist with detecting anomalies, flagging inconsistencies, and automating certain data-cleaning processes. However, AI cannot determine whether the underlying data is fundamentally sound or whether a particular data source is appropriate for a given use case. Human oversight and clear data governance remain essential.
—
## Conclusion
AI-powered development tools represent a meaningful leap in what product teams can achieve in terms of speed and output. For connected products, this acceleration is welcome but must be approached with clear eyes. The most consequential challenges—interoperability across systems, data trustworthiness, standards interpretation, and genuine user understanding—are not problems that code generation can solve.
The teams that will thrive in this new era are the ones that treat AI as a powerful accelerator for the work that already matters, not as a replacement for the judgment, empathy, and rigour that define great product work. Building connected products will always require both fast execution and deep insight, and the balance between the two is what separates functional products from exceptional ones.
Thank you for reading
