Every year we spend time looking at technology that isn't ready for enterprise yet. Not because we plan to use it tomorrow, but because understanding what's coming helps us make better decisions about what to build today. Some of this will matter. Most won't. Here's what caught our attention in early 2022.
What You Need to Know
- NLP has improved dramatically in the last two years, but enterprise applications are still limited by reliability and cost
- Computer vision is finding real applications in quality control and safety, particularly in manufacturing and construction
- Edge computing is genuinely useful for specific use cases (IoT, latency-sensitive operations) but over-hyped as a general trend
- Robotics is advancing faster in warehousing and logistics than in general-purpose applications
Natural Language Processing
This is the space moving fastest. Since GPT-3 launched in 2020, we've been watching NLP capabilities improve at a rate that's genuinely surprising. Text generation, summarisation, classification, translation, all of these have gotten meaningfully better.
The gap between "impressive demo" and "reliable enterprise tool" remains significant though. We've tested several NLP APIs for client projects and the results are good enough to be exciting but not consistent enough to be dependable. A system that's right 85% of the time sounds good until you realise it's wrong 15% of the time, and you can't predict which 15%.
The NLP models are getting better at generating text that sounds right. Nobody wants their customer communication system to be confidently wrong.
Mak Khan
Chief AI Officer
The cost model is also a barrier. API pricing for large language models is based on token volume. For a low-volume application, it's fine. For processing thousands of documents daily, the costs add up quickly.
Our assessment: NLP will become a significant enterprise tool within two to three years. Right now, it's useful for internal tools, prototyping, and low-stakes applications. High-stakes, customer-facing use is premature.
Computer Vision
More quietly than NLP, computer vision has reached genuine utility in specific domains. Quality inspection in manufacturing, safety monitoring on construction sites, inventory management in warehousing. These aren't research projects. They're deployed systems producing measurable value.
35%
reduction in quality control costs reported by manufacturers using computer vision inspection systems
Source: Deloitte Smart Factory Report, 2022
The common thread is that these applications have well-defined visual patterns, a controlled environment, and tolerance for occasional errors. A system that correctly identifies 95% of manufacturing defects is useful because a human reviews the flagged items anyway.
Where computer vision still struggles is in uncontrolled environments. Variable lighting, unexpected objects, novel situations. The technology works best when the visual domain is constrained and predictable.
For NZ enterprises, computer vision is most relevant in primary industries, manufacturing, and logistics. We're keeping an eye on applications in agriculture, where drone-based imaging for crop monitoring is getting increasingly practical.
Edge Computing
Edge computing, processing data closer to where it's generated rather than sending everything to the cloud, is one of those trends that's simultaneously real and over-hyped.
For specific use cases, it's genuinely valuable. IoT sensors that need millisecond response times. Remote locations with limited connectivity. Privacy-sensitive applications where data shouldn't leave the premises. In these scenarios, edge computing solves real problems.
The hype version, where edge replaces cloud for general workloads, isn't happening. Most enterprise applications don't need sub-100ms latency, and the operational complexity of managing distributed edge infrastructure is substantial. For the majority of what our clients build, cloud is fine.
Robotics and Automation
The warehouse robotics space has matured significantly. Amazon's acquisition of Kiva Systems back in 2012 kicked off a wave of investment that's now producing practical, cost-effective automation for logistics operations. Companies like Locus Robotics and Fetch Robotics are selling to mid-market warehouses, not just the giants.
General-purpose robotics is further out. Boston Dynamics' robots make great videos but the commercial applications are limited. Tesla's Optimus robot is ambitious but unproven. The economics of general-purpose robots don't work yet for most use cases.
Where we see the most near-term impact in NZ is agricultural robotics. GPS-guided equipment, automated harvesting for specific crops, and drone-based monitoring are all moving from experimental to practical. Given NZ's agricultural economy, this matters.
What We're Not Watching
A few things we've deliberately deprioritised.
Blockchain for enterprise. We've been sceptical for years and nothing in 2022 has changed our mind. The problems it solves for enterprise (trusted data sharing, audit trails) have simpler solutions.
The metaverse. Meta's investment is enormous but the enterprise applications are speculative. We'll revisit if something concrete emerges.
Quantum computing. Genuinely transformative when it arrives. Not arriving for enterprise use cases any time soon.
The Takeaway
The technology worth paying attention to is the technology that's finding real applications in specific domains, not the technology generating the most headlines. NLP, computer vision, and domain-specific robotics are all crossing the threshold from "interesting" to "useful." The timeline to enterprise-ready varies, but the direction is clear.
We'll do another one of these at the end of the year. With the pace things are moving, the landscape might look different by then.