I’ve spent many years in high-tech leadership, having worked in several product marketing and analyst relations roles. I’ve overseen rebranding campaigns, new product launches, and messaging strategies, working with many teams focused on delivering products that bridge multiple architectures and deployment strategies. My experience has been primarily from a perspective of enterprise software. This includes understanding buyer personas, value propositions, and how to achieve the lowest total cost of ownership. More recently, I’ve begun working in the hardware and chip development industry, in the area of AI semiconductor design. Interestingly, I see more parallels today between the software and hardware industries – a convergence I attribute to the increasing influence of AI.
Now the direction is to orchestrate AI capabilities across hardware and software platforms. This shift is an “agentic” transformation. This wasn’t always the case. In September 2024, Salesforce was an early pioneer in taking a software approach to building agents. One of their Dreamforce 2024 tracks, Build Innovative ISV Apps with Agentforce, discussed how “ISV partners can build Agents and actions for Agentforce using Apex, flows, and prompt templates.”
I don’t recall any conversations about taking an agentic strategy to hardware back in 2024, but I might have missed it! My focus at that time was on closely monitoring the field service management vertical of the enterprise software stack.
Today’s Reality
Now we are moving to a new world with the ability to bring compute, power efficiency, and edge-capable hardware closer to the demands of enabling next-gen applications, defined as agentic workflows – or software applications. My focus today is on launching a new Edge AI processing semiconductor chip (learn more about Newport, by ChipStart(R)). That position has given me a front-row seat to see the shifting boundaries between software and hardware in enterprise systems.
I see both worlds of hardware and software clearly. Each faces challenges today to remain viable, support the advances of AI, and deliver value. Software must overcome development and implementation hurdles. Hardware must continue to operate with greater efficiency and less power requirements. The best solution is to view these challenges as a collaborative, joint solution.
3 Ways AI Will Redesign How Enterprise Software Is Created, Coded, Implemented
1. Design via AI-informed requirement synthesis
AI tools will help gather and refine requirements automatically from user feedback, logs, and usage patterns. They might even propose new features based on predictive analytics that anticipate user needs before they are expressed. That means software specification phases will shift from human‐only driven to human + AI collaborative.
2. Code generation, augmentation, and optimization
We will increasingly use AI to generate scaffolding code, unit tests, sometimes even full modules. AI tools will optimize low-level routines for performance, energy usage, or parallelism. They will help catch subtle bugs, security vulnerabilities, or inefficiencies that humans routinely miss. Coders will move from writing every line to supervising, refining, and integrating AI-produced artifacts.
3. Implementation, deployment, and maintainability at scale
AI will change how software is deployed: think automated configuration, dynamic scaling, predictive failure handling. Monitoring and observability will use AI to interpret logs, trace anomalies, and suggest fixes. Maintenance cycles will shrink. AI will learn from past bugs, patch dependencies, anticipate obsolescence, and suggest refactors. Enterprise apps will evolve more continuously, with less human delay and more automated feedback.
Where Hardware and Software Are Merging—and What I’m Seeing
In my current role, launching a new Edge AI processing semiconductor chip, I encounter firsthand how hardware and software are converging. Here are several observations:
- Tight co-design becomes indispensable. To meet latency, power, and throughput goals, hardware design (microarchitecture, accelerators, memory hierarchy) and software (drivers, frameworks, AI models) must be designed together. One cannot optimize one without considering the other.
- Embedded intelligence moves to the edge. Rather than sending data back to central cloud servers, more processing can now happen in hardware near the source. That demands software that can operate under hardware constraints (energy, thermal, size) and hardware that supports flexible, rapidly evolving software.
- Model specialization and compilation. AI models increasingly must be compiled into formats optimized for specific hardware. Hardware vendors provide toolchains, compilers, SDKs. Software engineers must learn hardware constraints. Specialized accelerators for neural networks require software that translates high-level models into efficient low-level operations.
- Performance/per watt becomes first-class metric. In enterprise settings, it’s no longer enough for software to run fast. It must run with predictable performance, minimal power consumption, and acceptable thermal profiles. Hardware innovation (e.g. semantic sparsity, novel memory designs) enables software performance that meets real-world deployment constraints.
From observing current product roadmaps, I see hardware feature‐requests being driven by software model demands and AI algorithm behavior. Equally, software frameworks are being built to exploit hardware capabilities – tensor cores, specialized units, quantization support. The two used to be loosely coupled; now they are merging.
The Future: Uncertain, But Not as Most Imagine
Of course, the future is still largely unknown. Especially today, with so many rapid advances and changes! Many assume we will extend today’s trajectories: bigger models, more cloud compute, more centralized data centers, incremental improvements. But I see things diverging.
We may see radical decentralization of compute: more intelligence at the edge, inside devices, sensors, appliances. I realize this isn’t a new concept, but one that has been touted for years. But now, I believe its time has come. Software may need to be designed with greater focus for intermittent connectivity, for energy scarcity, for heterogeneous hardware everywhere.
Perhaps AI tools will automate much of what we now consider craftsmanship in coding and system design. I imagine a world where designing new enterprise software becomes more about defining intent, verifying behavior, ensuring safety, rather than writing every module by hand.
Hardware may evolve in unexpected directions – neuromorphic designs, photon-based chips, very low precision computing – that force software to adapt, perhaps to rethink algorithms or accept trade-offs unfamiliar to current teams.
One thing is certain. The way most people envision AI’s impact is too narrow. They focus on models that are bigger, cloud-only, centralized, or purely software. But the systems that win will be those that unify hardware + software in ways that manage energy, latency, trust, and edge constraints.
Closing Thoughts
Enterprise software and enterprise hardware are no longer separate silos. AI forces their alignment. Based on my career trajectory and current role, I see convergence in design, coding, and deployment. The future will demand collaboration between hardware engineers, AI modelers, and software architects in tighter loops.
What the future holds is still open. But if we stay anchored to old assumptions, we risk being surprised. An open mind is not optional. It’s essential. The big shifts will come where we least expect them. Let’s be ready.
