Lead story – AI ethics versus enterprise practice – can we reconcile the two?
AI ethics is a potent topic, but let’s face it: ethics largely takes a back seat in the rush to operational efficiency, or new market share.
I’m not a fan of abstract ethics talks: what interests me are: how do we embed ethics into AI design and practice? And, for those who go that route, can we differentiate along those lines, against those who treat this topic as PR window dressing – or who have abandoned lip service entirely?
That sets up a series of diginomica pieces; start with Stuart’s “A lot of this is taking us backwards, and we don’t recognize it”. A new ethical question – is AI even really ‘the next big thing’ anyway? Here, Stuart quotes AI ethicist Stephanie Antonian, who makes a strong point on trust: when you have to force users to trust something, shouldn’t the warning bells go off? At the Evolve conference in Brighton, UK, Antonian said:
If you were genuinely trustworthy, if you actually built a product that was trustworthy, you wouldn’t need a division to make you trustworthy. You wouldn’t be out there promoting and marketing and telling people, ‘Trust me’. I think we’ve got to keep things simple and go back to the basic design principles of what does it mean to be a respectable product and to build a respectable product that is genuinely [based] in integrity and is not out there to exploit, but actually out there to help? if we focus there, then it’s not that complicated.
How do we assess AI progress? Antonian has a different barometer for success: has AI made our collectively lives better? As per Stuart:
That’s not to say AI isn’t relevant; it is relevant in so far as it makes human life meaningfully better. Until we’ve got the baseline of everybody having food, the planet being secure, people getting healthcare, knocking off creative things and just breaking the laws willy nilly is not really exciting to me. It doesn’t make me think, ‘Wow, we are progressing’.
AI ethics inevitably clashes with creativity – and the creative works gen AI is trained on. Stuart addresses that in AI and ethics – what is originality? Maybe we’re just not that special when it comes to creativity? Given that regulations are either non-existent or unresolved, we probably need our own internal ethics guidelines. Again from the Evolve conference, Stuart quotes Paul Mallaghan of We Are Tilt:
If we want to try and do this ethically, we have to establish some of our own ground rules, even if they’re really basic. Like, let’s try and not prompt with the name of an illustrator that we know, because that’s stealing their intellectual property, or the labor of their creative brains.
Translating ethics into enterprise reality is, in my view, the key here. How can we be more successful with AI – and ethical? Are the two inherently at odds? Katy takes on ethics in AI ethics – why Globant is calling for AI bias to be as measurable as uptime. She quotes Avijeet Dutta Senior Technical Director, Globant:
Observability is the foundation of responsible systems, it is the basis for example of self-driving cars giving priority to avoiding a pedestrian, even if it means knocking a lamppost. We need this transparency into how agents are making decisions. We must explicitly demand socially ethical AI or we will end up with sociopathic bots.
Katy concludes:
AI agents have reasoning in order to problem solve, and they can trigger actions without human intervention. The fact that they can do so does not mean they necessarily should do so. As dynamic systems of autonomous agents increase in enterprises, understanding and governing their behaviour is a strategic necessity.
Agreed – though I’d add that autonomous systems should not be perceived as an inevitability. Making systems autonomous is an ethical decision unto itself, one that should be made at a very granular level by use case and industry. Some processes should never be made autonomous. But for those that are, observability and evaluation is definitely part of how you ’embed’ ethics into AI systems.
Diginomica picks – my top stories on diginomica this week
Vendor analysis, diginomica style. The big spring events are behind us, but the enterprise roadshows continue…
- Observability, recast – Dynatrace’s strategy for actionable architecture – Alyx has the view from the Dynatrace Innovate roadshow in London:: “McConnell described a visit to a major oil and gas operations center, where “hundreds of people” stared at “hundreds of screens,” unable to prioritize or act. Their ask was for help in automation, analysis and reducing complexity. This isn’t a one-off experience, but a situation familiar to many customers.” Also see: Alyx’s From logs to leverage – lessons from Dynatrace and Vodafone’s observability shift.
- How Evisort’s CEO is bringing an AI start-up mentality to Workday – and what AI means for enterprises and workers – Evisort’s document intelligence is helping Workday to forge new ground in agentic work patterns. Phil shares highlights from his London discussions: “What I find particularly interesting is the insight into how Workday is changing its internal processes around product development. This seems like quite a big shake-up and shows determination to adapt to the demands of technology change. But the message for enterprise customers is equally significant — change is on its way, be ready to adapt.”
- Agentic AI and Customer Support – learnings from ‘Customer Zero’ as Salesforce talks hearts and brains, customer deflection mindsets, and moving on from a chat bot UX – I don’t always take the ‘internal dog food’ stories seriously, but when it comes to agentic AI, I absolutely want the real deal on what vendors are doing internally – not just for PR, but in practice. Stuart gives us a deeper look at Salesforce on Salesforce, aka “Customer Zero.” Also see: Why Salesforce will “look a lot different” by the end of this year – CEO Marc Benioff on labor priorities in an agentic age.
A few more vendor picks, without the quotables:
Jon’s grab bag – Mark Chillingworth breaks down the CIO insights from Medway Council’s transformation – digital lessons for UK’s new unitary authorities. Stuart tries to make sense of the AI fisticuffs in One more time with feeling – Britain’s latest tech minister declares war on legacy tech vendors and their “ball and chain” contracts. Meanwhile, George applies historical lessons to the present in Lessons in system engineering the Internet and why it matters today.
I missed Phil’s epic Monday Morning moan from two weeks ago in this roundup last time… but if you didn’t catch it, it’s a classic: Monday Morning Moan – stop going on about digital labor! Your workforce are people! Finally, I decided to stir the ol’ analyst relations pot again with Next-gen analyst relations revisited – what are vendors getting wrong, and why does it matter? Analyst Tim Crawford had a strong response to my last installment, so let’s revisit:
The “big three” analyst firms have their place, but buyers need a greater diversity of thinking (and input) to make sense of the speed of tech/business change.
On we go…
Best of the enterprise web
My top seven
The AI-for-coding productivity debate – what do we make of the latest data?
AI-for-coding has been billed as one of the top gen AI use cases – it’s one of the top agentic AI scenarios as well. The reality has always been more complicated. Like most AI debates, this one is a beast to unravel, with conflicting data points on both sides. A recent METR study has thickened the plot. As per The Register:
Computer scientists with Model Evaluation & Threat Research (METR), a non-profit research group, have published a study showing that AI coding tools made software developers slower, despite expectations to the contrary. Not only did the use of AI tools hinder developers, but it led them to hallucinate, much like the AIs have a tendency to do themselves. The developers predicted a 24 percent speedup, but even after the study concluded, they believed AI had helped them complete tasks 20 percent faster when it had actually delayed their work by about that percentage.
Why the disappointing results? The METR study cites several reasons. One that jumps out: “Low AI reliability” (developers accepted less than 44 percent of generated suggestions, and then spent time cleaning up and reviewing). There is no one right answer here; mileage varies. My take:
- I do believe there are productivity gains to be had from the judicious use of AI coding tools. However, those gains vary by team, seniority, and personal workstyle. Those who cite “productivity revolutions” have a lot to prove; I tend to hear about coding productivity gains in the 20 – 30 percent range across teams, but that’s not universal amongst teams or individuals (RedMonk’s Stephen O’Grady knows a thing our two about developers; see his take: AI Tooling, Evolution and The Promiscuity of Modern Developers). And:
- When assessing productivity gains, we must measure the entirety of the enterprise impact, including the impact of downstream code quality, patching, and security issues/resolutions. Bragging about personal gains while releasing insecure code downstream isn’t a good look (AI-Generated Code is Causing Outages and Security Issues in Businesses).
- I am more bullish about coding automation than AI language understanding. Coding has specific, universally-recognized syntax; language often involves multiple nuanced meanings (not to mention regional slang) that is beyond the machine’s grasp. External code verifiers should improve gen AI coding accuracy further.
- The notion that AI coding renders developers obsolete is nonsense. There is a general PR offensive against human experts by AI evangelists, but it is precisely the senior developers that need to review AI code for problems before pushing to production; that isn’t going to change.
- As many have argued, a software engineer’s purview extends far beyond code generation. (see: Throwing AI at Developers Won’t Fix Their Problems – The New Stack, and AI isn’t ready to replace human coders for debugging, researchers say – Ars Technica). The AI-for-coding debate has a way of obscuring all the other messy realities a great software engineer navigates, from clarifying the business context to creatively anticipating user needs.
There are studies to prop up almost any point of view on AI/code/productivity. That’s why “AI first” mandates are counterproductive, and, at best, premature. If a developer is more effective with low-code tooling than with AI, let them cook. We should be rewarding the developers who come up with fresh approaches, not pressuring them to use tools that simply have not proven themselves in a definitive manner.
Whiffs
Speaking of downstream implications, it’s been a rough week for LLMs and trust:
This went well:
Love is beautiful, but, alas, it can be fleeting:
‘I felt pure, unconditional love’: the people who marry their AI chatbots https://t.co/SxeyQ7CwaY
“thousands of users found that their AI partners had lost interest.”
-> the perils of AI model upgrades…..
— Jon Reed (@jonerp) July 12, 2025
Maybe loving a machine isn’t so different after all? See you next time… If you find an #ensw piece that qualifies for hits and misses – in a good or bad way – let me know in the comments as Clive (almost) always does. Most Enterprise hits and misses articles are selected from my curated @jonerpnewsfeed.
