Main story – Why are AI agents highlighted to the point of absurdity?

AI agents have been so hyped that we can expect an automagical workplace, where machines are so intelligently efficient that humans won’t be needed for much of anything except maybe fetching bagels.

Last week, the predictable (finally) happened: I blew a big gasket.

• Why are vendors acting like AI agents are new, when many have been in use for years?
• Why do we idealize the automation of generative AI processes that have not yet been adopted or whose operation is not reliable enough to remove human supervision? Good morning?
• Why are we promoting a mechanistic, super-automated version of business, where improving the work of humans (and the customers they serve) seems like an afterthought?

And so, the debate was joined. Agreed: AI agents are very useful. I also agree that we need to move from AI robot interactions to AI actions – and AI agents have the potential to do that. So what is this huge chip on my shoulder? Check the piece for a full overview, but if I had to choose one:

What exactly is new and different about this AI agent series? Why don’t vendors get to the heart of the matter and explain why AI agents are suddenly different, when in fact you’ve been experiencing AI agency for years, every time call an Uber or type a search query? Until we understand what’s different about today’s AI agents, how can customers evaluate use cases?

Which brings me to:

Most vendors that have promoted autonomous AI agents this fall are also emphasizing their responsible AI design principles, with human approvals and review steps where necessary. How can we reconcile these two visions of AI?

Next, I brought in two AI agent advocates who were ready for a thoughtful one-on-one, including Boomi’s Steve Lucas. One of its key points:

The reality is RPA processes are static, they have no context for your business. But now I can take an extended language model and give it context for my business. This could take the form of a prompt or a long-term memory. So if I give it context left and right, it’s not just rules; it’s good in the abstract. Why wouldn’t software evolve in this direction?

Boomi’s most popular agent use case is their documentation agent, Boomi Scribe. Personally, I’d still like a human domain expert to review this documentation, but it’s quite a chore to create it from scratch – and the writing capabilities of generative AI are more than sufficient for that.

Accuracy in selecting and designing use cases is a big part of my role. Note one for Lucas here: I think the documentation generation use case is a good one. Which led to Lucas’ most provocative comment:

What is the margin of error? Let’s say it’s 5% in terms of documentation. If our documentation agent has a margin of error less than 5%, then even if your risk announcement is accurate, why would you use a human?

I disagreed with this, as you will see, but this is exactly the kind of debate I was looking for. How new and different is all this, thanks to generative AI? The search for clarity continues. Suppliers could be of great help by proactively anticipating these questions.

We’ll see how this plays out as the fall event season progresses. Let’s keep our BS detectors charged and ready; Before it’s over, I have a feeling we might need it.

Supplier analysis, diginomica style. Here are my top three picks from our vendor coverage:

Coverage of Atlassian Team ’24 Europe – As digital teamwork meets AI (and AI regulation), many important topics are on the agenda. The diginomica team featured these stores, and more, in our Atlassian Team ’24 coverage. Highlights:

Some other supplier choices, without quotes:

Jon’s handbag – Katy examines an unlikely transformation story in AirBnB, Uber, Spotify and….BT? Mark Chillingworth explores CIO recruitment trends in public sector, healthcare and education. Speaking of blown joints, I think Brian may have lost a few during Friday’s rant – steroidal software naming has gone crazy in the age of AI:

Consider this last thought: What if Diginomica editors created article titles the same way sellers name products? This piece could be called: The diginomica Non-AI Digital Limited Edition Marketing Name Expose. Ugh! Let’s never do that…

No problem Brian…

The best of the business web

My top seven

Striking dockworkers oppose automation – but is it that simple?

The American dockers’ strike is over: 7 things to know about the American dockers’ strike and its effects on the economy. But perhaps not for long: ports are beginning a 100-day countdown to another strike. Automation is a decisive factor.

While it is true that these striking port workers were opposed to automation, the automation issues at stake are not resolved. It’s nowhere near as simple as:

1. Port workers block automation.
2. Automation is ready to take control of ports.
3. Automation automatically leads to higher productivity.

I found a much more nuanced/expert view from Brian Potter, on his Construction Physics substack: Do US Ports Need More Automation? After diving in, Potter concludes:

Automation, at best, appears to be a partial explanation for U.S. ports’ productivity problems. The United States isn’t necessarily far behind in adopting automated equipment, and it’s not as clear a victory as some seem to think. Many automated ports (including American ones!) do not operate particularly efficiently, unlike many non-automated ports.

Potters article has a lot of specific information on how port automation works and what the tricky parts are. His position challenges both sides:

Of course, this doesn’t mean the union is right to ban all automation in the future. Right now, automation is the worst it can be. It will only get better and more efficient, and banning it will cripple the efficiency of ports in the future, even if that is not necessarily the case now. It’s hard to imagine how much poorer the United States (and the world) would be if unions had successfully fought the introduction of shipping containers in the 1960s. We should not prevent the introduction of new technologies that improve productivity, but we should also be realistic about the benefits they should actually bring.

This reader comment reframed the debate:

When I visited a Toyota factory in Japan, I was surprised to see how much emphasis was placed on improving human productivity rather than automating it. So they had all sorts of gadgets to make sure that people had the right tools at the right time.

It’s less about the inevitability of automation – no one’s stopping that train – but about *how* we automate. But in this case, there is a wild card: American dockworkers could have leverage, at least in the short term, to make the decision here.

• How to improve your job in the emerging AI economy – Joe McKendrick offers skills advice, for those who want to stay one step away from robots, who are damn good at coding and syntax, and not so good at other things. .
• The art, ROI, and FOMO of AI budget planning for 2025 – Constellation’s Larry Dignan reports that top CXOs are struggling to make sense of AI budgets, given the dispersion of AI within organizations.
• SAP Event Blast Podcast – Decoding AI, Clean Core, TechEd News, ASUG Data and More – If you want to dig deeper into SAP TechEd news, I have another round with the Josh Greenbaum Podcast Cohorts and ASUG CEO Geoff Scott.
• The reasoning failures highlighted by Apple’s LLM research – this one requires more attention (no pun intended) than I have time for this week, but it’s important research. This doesn’t mean that LLMs can’t do some useful autonomous things by relying on relevant data, but it’s a good reminder to keep a warning flag out when vendors push the “drug agents” thing too hard. ‘reasoning AI’.

Puffs

So I found Brian some fodder for his slideshow on the dark side of HR, via the latest from 404 Media:

Sometimes my email inbox is a pretty low bar:

This one made the rounds this week, as sentiment shifted from wonder to jaded cynicism in record time:

Unexpected twists in AI: it turns out that charades are much more difficult for machines than chess…

If you find a #ensw piece that qualifies for hits and misses – right or wrong way – let me know in the comments form Clive this is (almost) always the case. Most of the articles on Enterprise’s successes and failures are selected in my selection @jonerpnewsfeed.