Software Oasis

Your company already has a shape. The cost of human labour set it. Departments, queues, approvals and sampling are not laws of nature, but most AI programmes treat them as such. They automate the constraints they should remove. Once work can scale with compute, the question is no longer how much AI to add to the company you have. It is whether you are prepared to build the company that could replace it — and own it before someone else does. The final article in my Digital Labour series: https://lnkd.in/dZhRuzXW Digital Workforce Services Plc the home of your digital labour Agent Workforce

Excited to be talking about the consequences that our enterprise decisions are having on our kids and other vulnerable populations at Black Hat. Sign up, hang out, join the conversation.

Every week there's a new AI breakthrough. Every week there's a new question about what it actually means for the people building and running enterprise systems. And every week, the gap between the hype and the reality gets a little wider.   We're launching the Nextworld Tech Blog to help close that gap. Our engineers have spent years building AI into enterprise software. They have hard-won perspective on what works, what doesn't, and what's coming next. Now they have a place to share it.   My first post covers what to expect and why we think this matters right now. I hope you'll come along for the ride. https://lnkd.in/g64jKUqm

Big moment for the Worth AI team: we've raised $30M in Series A funding to accelerate our mission to modernize how financial institutions onboard and underwrite customers. This round was led by Fulcrum Equity Partners, with participation from Amex Ventures and TTV Capital. For decades, institutions have relied on fragmented systems not built for speed, intelligence, or scale. That model is breaking, and Worth AI is building what comes next. Incredibly grateful for our wonderful team members, partners, and customers around the world! Read more from Inc. Magazine: https://lnkd.in/eZH8AN6d

Many IT and security teams are deploying AI agents using the “F* Around and Find Out**” methodology. The pressure is real. CEOs want AI deployed yesterday, and many technology teams are responding by rolling out agents in “YOLO mode” just to keep up. The industry’s first response has been agent discovery and observability. That’s a good start. But it’s NOT enough. Knowing which agents exist and watching what they do doesn’t answer the questions that actually matter: 🚧 Should this agent exist? 🚧 Who owns it? 🚧 What is it allowed to do? 🚧 What data can it access? 🚧 What tools can it use? 🚧 Who approved those permissions? Observability tells you what happened. Organizations also need the controls to determine what should be allowed to happen. The companies that succeed with AI won’t just have visibility into their agents. They’ll have governance, policy, approvals, and accountability built into every agent from day one.

Okay, let's talk semantic debt. One of the biggest misconceptions in data + AI right now is that better models will solve agentic analytics. They won’t, at least not by themselves. What we’re seeing is that teams hit the wall much earlier. The agent can write SQL, suggest a dbt change, and even trace part of a failure. But production usefulness breaks on a different layer entirely: semantic debt. That debt shows up as inconsistent metric definitions, undocumented views, one-off schemas, partial lineage, tribal knowledge around transformations, and no clean way to prove why a result should be trusted. This is why so many “data agent” demos look impressive and then get hard to operationalize. The issue is rarely generation alone. Getting the agents to produce answers is super easy. It is whether the agent has enough grounded context to answer questions like: - Which definition of this metric is canonical? - What breaks downstream if I change this model? - Is this dashboard number wrong, or is the pipeline wrong? - Was this table created through a governed workflow or by someone experimenting on Friday night? - What evidence should a reviewer see before approving the change? A few recent signals all point in the same direction. dbt’s Developer Agent preview is a real sign that analytics engineering is becoming an agent surface. Their more recent post on the “semantic debt crisis” gets even closer to the heart of it: when AI scales access to data work, it also scales inconsistency unless the underlying semantics are governed. Anthropic’s engineering work on harness design makes the same point from another angle. Better results come from better scaffolding, better evaluation, and better operating constraints around the model. My take: the next generation of winning data agents will not be defined by who generates the prettiest SQL. They’ll be defined by who can combine semantic context, lineage, provenance, execution history, and reviewable evidence into a system that data teams can actually trust. That is where the real leverage is. And it is also where the real product gap still is for most of the market.