Enterprise hits and misses - agentic AI project failure versus success, open source versus AI, and the perils of disconnected CX

Lead story - Getting real about agentic AI projects - what's working, and what's not?

My editorial self-challenge: boil down a year's worth of research/dialogue on AI project success (or lack thereof) into one post. And do it for an audience that must make sense of wildly disparate (and at times goofy) daily narratives: 

What is an executive to make of so many conflicting headlines? AI is supposedly transforming industries, but AWS is knocking systems offline with vibe coding. We are in dire need of jugular context that weighs the upside and the downside. 

Set aside the over-hyped bluster of failed/underwhelming enterprise tech (blockchain, metaverse). In past technology surges that actually mattered to culture and commerce (web, mobile), the enterprise was unquestionably a laggard. But in AI, the enterprise may lag at times, but the good ol' risk-managed enterprise mentality is exactly what good AI projects need. To that end, I single out three crucial enterprise AI advances: 

• Context at the time of inference - getting AI systems the best information at any moment in time, for that particular company and user, in a well-governed way. (A work in progress, but the right approach).
• Constraining LLMs in a "compound systems" architecture, combined with other forms of machine learning, along with deterministic systems, and external tool calls to verifiers, rules-based automation, or sources of database truth.
• A promising distinction between off-the-shelf frontier models and domain-specific/smaller models, informed by relevant data.

The rest of the post is a spicy review of what's working in agentic AI, and what's not. What's not working? 

1. "AI First" -  If you have to impose your AI tools on your workforce, something is wrong with your tools.
2. Agentic for the sake of agentic - go figure: picking the right tool for the job remains the best idea, especially when the shiny new tool is expensive and probabilistic, as Walmart just learned the hard way (Walmart fires OpenAI in playbook-changing move).
3. Multi-agent protocols - "Why are we hyping agent-to-agent science experiments when customers need wins now?" (Yes, there are exceptions, which I get into).
4. Layering agents onto bad processes, crappy data and custom code - captain obvious nomination, but it had to be said - AI is icing on your architectural cake.
5. Reckoning with AI workslop and downstream impacts - AI workslop isn't working. Neither is measuring the upstream ROI without reckoning the downstream impact (see: AWS vibe code outage).

What is working? It's a list with some surprises: 

1. Granular autonomy - giving customers the level of AI autonomy they want, and the ability to dial it it up and down at a per-process level.

2. Evaluation and observability.- if vendors provide the tooling, and customers apply it. 

3. Explainability - "explainability" via LLM reasoning is overrated, but explainability via proper data context (and source links) is a thing.

4. Use case design - You can have two identical customer service agentic AI deployments - even using the same technology - and one can succeed, while the other fails. It's all in the design, and savvy human escalations etc.

5. AI readiness - "AI readiness means acting on the organizational implications of proper use of AI. How? By breaking down data silos - and building cross-departmental teams tasked with rethinking processes, and establishing governance frameworks." Oh, and stacking up wins along the way - examples noted in the piece. 

6. AI for data quality/cleansing/metadata annotations - a work in progress for sure, but a promising area for making data projects more palatable.

7. Good old fashioned project discipline - Such as? Establishing systems of accountability,  maturity models to track your evolution, and customer-driven KPIs.

I hope I've lured you in to check the full argument - there are more than thirty resource links to dig into throughout the article, if/when you need supporting context.

The enterprise AI argument in a nutshell? Modest AI success is sexy, and for now, it's enough. Oh, and: set your own internal tone. You don't have to let AI escalate employee fears and data surveillance. AI can be about business model experimentation and growth, rather than new turbo-powered, Darwinian worker hamster wheels. Operational efficiency goes a lot better when you see the upside, rather than your career circling the bowl. And: unlike OpenAi, enterprises don't have to justify a $1 trillion infrastructure commitment: 

Modest successes add up, even as markets demand exponentialism and dreamy 10x productivity gains. How about we start with repeatable results instead, and go from there? 

Diginomica picks - my top stories on diginomica this week

• the diginomica network pulse - what CIOs actually think about AI and the SaaSpocalypse - Mark Chillingworth shares a notable research update from diginomica research, on what else - the topic du jour. I'll dig into this one further for you next week, but for now: "Our pulse survey reveals AI and SaaS will become part of the enterprise technology mix. For providers of SaaS and AI, that means a dogged fight for some profit margins from CIOs and CTOs with a plethora of competing demands and business challenges."
• In pursuit of the customer Holy Grail - why organizations must finally tackle the cost of dis-connected CX - Rebecca releases some significant new research from Valoir on CRM data obstacles, and their AI impacts: "The next phase of AI in customer service won’t be defined by better models or flashier demos. It will be defined by who gets their data house in order, and who doesn’t. We’re already seeing the early signs of a shift. Leading organizations are moving away from brittle, integration-heavy architectures toward platform consolidation, unified data models, and shared data layers. They’re prioritizing real-time access to consistent, contextual data across the enterprise."
• E-Bomb! US e-commerce is the preserve of the old and rich, warns new report, while most of the world skews young  - Chris delves into some surprising e-commerce findings: With the notable exceptions of the US and Europe, this new report finds: "The vast majority of the world's online buyers are younger, earn less, and pay differently. They are entering the digital marketplace for the first time, and they are doing so on their own terms."
• Why Karl Friston is betting on cultivating curiosity for sustainable AGI - When AI/active inference pioneer Karl Friston talks, I listen. George's interview shows why. Friston is tackling the true frontier questions of AI, while his company, VERSES, looks to productize them. The ARC-AGI-3 challenge underway now will be fascinating... Also see: George's Why the boundary problem is the biggest challenge in agentic AI. 

Vendor analysis, diginomica style. Here's my three top choices from our vendor coverage:

• The big bets are on as Salesforce pitches the need for enterprise transition from model to system level AI - Stuart on how Salesforce's AI Foundry initiative will look to tackle the issues needed to achieve "system-level AI."
• Oracle's agentic Fusion play - from system of record to system of outcomes - Oracle issued some big AI agent news last week, with 22 new agentic applications. Here's Derek's on-the-ground take from London: "For buyers, this is well beyond the capabilities of copilots and generative AI assistants. Those tools may have provided a marginal uplift in productivity (and even that is debatable), but this suite of agentic applications essentially changes how work gets done."
• Sage finds a vigorous pulse among SMBs in the UK - but signs of caution too - Phil delves into fresh SMB data from Sage, via a recent London event. On Sage's new quarterly (and monthly) customer index, Phil writes: "Being able to tap into these more up-to-date metrics is particularly important at a time when unexpected events such as the recent conflict in the Persian Gulf can rapidly change the economic landscape at home."

KubeCon Europe 2026 - Alyx was on the ground in Amsterdam, where notable breaking news mixed with forward thinking on where observability is headed: 

This roundtable discussion felt like a real turning point. When Lewis described observability as "a platform that's going to inform business growth" and Fajerski insisted it is "a property of a system, not something you do," they were not disagreeing. They were describing two halves of the same transition: from a technical function to a business capability, from cost center to value driver. The AI agent question is the accelerant.

• Observability is becoming a business language problem - and KubeCon's practitioners are re-writing the dictionary
• OpenSSF's CRob on why open source security is still a people problem - and why AI is making it worse before it makes it better
• Kubernetes puts ingress nginx to rest at KubeCon - 'Nobody can keep it safe'

A few more vendor picks, without the quotables:

• Zendesk brings on AI-native agentic CX capabilities with its acquisition of Forethought - Phil
• Why Genpact is warning about AI driving the mass-generation of technical debt - Katy
• Celonis - Europe's defense reckoning has an execution gap and a sovereignty problem - Derek
• As customer journeys fragment across AI and chat, Contentsquare adds new analytics tools. Here’s what’s new - Barb

Jon's grab bag - Mark Chillingworth visits a groundbreaking new digital skills college in Ada, the college bridging the digital skills gap - and giving young people a fighting chance. Stuart curated another batch of newsworthy bits in The long and the short of IT - the week in digibytes, including Meta's no-good-very-bad-week in court. 

No one better to weigh in on this issue than Phil, who analyzed the birth of SaaS, and now we come full circle: The AI-native naysayers are making the same failed objections we heard to cloud-native apps two decades ago (I am going to reign in my take on this in, but Phil's piece is well worth a read. I'll just say that while AI and SaaS are in some (useful) conflict, they are also much more complimentary than SaaS versus on-prem... No better way to build enterprise-grade AI apps than on a (true) SaaS apps/data platform). 

If Anthropic's legal drama is your thing, Stuart has it for you blow-by-blow. The latest? First blood to Anthropic as US judge slams Trump 2.0's “Orwellian" attempt to cripple the firm as unconstitutional and illegal. Madeline ties the inclusion and AI themes together in a must-read piece: AI needs to be inclusive by design – here’s how the NHS, Microsoft and GoFibre think it can be done. Does AI and music have an upside? No one better than Chris to assess that: The Universal – has music’s AI future finally been sold? UMG CEO sets out his AI vision. 

We wrap the diginomica content binge with Alyx's in-person Monkigras review: 

In a moment when institutional trust is declining, political environments are hostile, and the open source ecosystem is under genuine strain, a conference that centers human connection, psychological safety and community resilience isn't a nice-to-have. It's a model for what the industry needs more of. That's what the open in open source is supposed to mean. (Monki Gras 2026 - the most important thing in open source isn't the code). 

Best of the enterprise web

My top seven

• AI Forum 2026: There isn’t an easy button for AI  - Constellation's Larry Dignan on their recent AI executive and AI forums: "Enterprises need to focus on the business logic and deterministic workflows more than non-deterministic agents that'll allegedly just figure it out... The AI control plane needs to be infused with context that can improve agent accuracy and enable them to scale."
• 96% of codebases rely on open source, and AI slop is putting them at risk - "AI-generated slop is overwhelming open source maintainers with low-quality pull requests. Here's how projects are fighting back with policies and new tools."
• The Contact Center Is Dead: Long Live the Operations Layer - Another strong summary of a CRM Konvos discussion by Thomas Wieberneit: "Let’s just be absolutely clear from the start: nobody wants a ticket. A ticket is simply a formalized receipt of failure. It is documented proof that a product broke, a service failed, or a user interface was too clunky to navigate."
• Anthropic and OpenAI just exposed SAST's structural blind spot with free tools - This Louis Columbus piece is a couple weeks old, but it is worth a close look. These tools definitely seem useful, but I'll double down on this: "Neither Claude Code Security nor Codex Security replaces your existing stack."
• Meet the New Dr. No. - Lora Cecere on the new supply chain decision maker, who definitely knows how to say no: "The deployment of new approaches requires the learning of a new language, rethinking first principles, and being open to new outcomes, all of which fly in the face of tradition."
• SAP acquires Reltio to build out SAP Business Data Cloud - Larry Dignan is back, this time to go inside a data-related SAP acquisition.
• Multi-media picks: Fresh off a spicy Sage analyst day, usual suspect Brian Sommer and I taped a podcast (until the vacuum cleaners shut us down): On ERP value, AI disruptions and where we go from here - hot issues from Sage Analyst Day 2026. Also, check this YouTube pick which deconstructs the "$1 trillion AI context graphs opportunity" post: Context Graph and Process Knowledge, Jessica Talisman, Contextually LLC.

Whiffs

Meta's not alone... OpenAI is having a fun week: 

The perils of move-fast-because-it-sounded-cool: 

Buzzfeed may be a punch line now, but it was once a groundbreaker. Is there a worse-example of drunk-on-AI "AI first" than this corporate (mis)adventure? 

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.