The transition into delivery is a change program in its own right One recurring failure point is the transition from discovery into delivery. Discovery creates insight, decisions and possible solutions. Delivery requires a shared operating rhythm, clear ownership, usable backlogs, agreed evidence, realistic capacity and a way to handle what is still unknown. The assumption that teams will naturally move from one mode to the other is dangerous. A deliberate transition should make the first delivery period explicit: What outcome is being pursued? What is the smallest valuable slice? What is…
Delegation will also vary across the strategy-to-execution chain. Identifying signals, forming hypotheses, prioritizing investments and executing initiatives involve different levels of ambiguity and consequence, so they should not inherit the same human-agent decision rights. Recent research from MIT's CISR offers a genuinely useful lens here: decisions should be split between human and AI based on two dimensions, i.e., ambiguity and risk, and the split isn't binary but plays out across three activities: framing a decision, acting on it and learning from the outcome. Framing: Deciding what's…
The way forward: Funding the infrastructure that makes autonomy scalable The controls around an AI agent aren’t a brake on adoption; they’re the infrastructure that lets an enterprise pursue the benefits of agentic AI with confidence, resilience and commercial defensibility. The Agentic Scope of Authority Framework turns that idea into decisions leaders can fund, assign and audit. The basic premise is simple: authority must be clear inside the enterprise and credible to everyone outside it. Actual authority is what people authorize and the architecture enforces: the task, limits, tools, data,…
That loss compounds every month a company keeps paying for infrastructure that's technically working but practically idle, the AI equivalent of leaving the lights on in an empty office. But that's not the whole problem. The bigger loss never shows up as a cost at all; it shows up as revenue that was supposed to happen and simply doesn't. For every insight a company generates but never acts on, the real bill isn't the price of the model. It's the price of everything that model was meant to protect or unlock, quietly not happening. Why insight stalls before it becomes action Typically, the…
In part 1 of the AIRD anthology, our focus was on foundational prerequisites for an AI-ready platform. This included ensuring the data fueling the platform was high quality and governed securely so it could be trusted; using metadata and cataloguing to improve efficiency across a data-first organization; and establishing a semantic business layer so AI systems could identify and use the right dataset for each task. Once a solid foundation is built, more sophisticated components and topics can be introduced. This is the overarching focus of Part two. A to B accessibility As this series…
After generation, use a yes-or-no for each claim. Split the LLM's answer into individual claims. A simple sentence splitter is usually enough. For each claim, ask whether the retrieved chunks support it, then flag or remove anything below your threshold before the answer reaches the user. Why Jev? A grounding check only helps if you run it on every answer, and that's only realistic when each check is fast and cheap. In every case, Jev will deliver a probability and an answer. That's what makes the next part possible. Use the probability, not just the answer The real value of Jev here isn't…
The missing link in autonomous decision-making Your AI agent can have access to the right data, produce a perfectly plausible answer and still make a terrible decision. This is the emerging risk of agentic AI: the context problem. As your organization delegates more complex work to autonomous systems, the growing challenge is whether your agent understands what that information means inside your business: which definition applies, which exception matters, whose judgment to trust and when a rule should be broken. Put simply, context is the knowledge that makes AI trustworthy. While the…
Warehouse, lake or lakehouse: How to pick Data warehouse: structured, schema-on-write, built for business intelligence (BI) and reporting with mature SQL tooling. A good fit when your workloads are well-defined, mostly structured and governance/consistency matter more than raw flexibility. Data lake: cheap storage, schema-on-read, holds structured, semi-structured and unstructured data side by side. Great for data science and exploratory work, but without discipline, it turns into a data swamp nobody trusts. Lakehouse: the industry's answer to not wanting to run and reconcile two separate…
In one project, a product manager built much of a prototype directly, shortening a feedback cycle that would traditionally have required coordination across product, design and development. The result is not that everyone becomes an expert in everything, but that the boundaries between roles become more permeable. 3. From document to managed context: Documentation remains important, but AI creates a new requirement: organizational knowledge must also be structured and accessible as usable context. On some AI-enabled projects, requirements, specifications and architectural decision records are…
For more than a decade, the default response to a new business requirement was often to add another application or expand an existing suite. This was fast and convenient, but it came with a trade-off. Instead of the software adapting to how the organization worked, organizations were frequently forced to change their workflows to fit the software. While this approach may work well for a commodity process, it becomes a constraint when the workflow is core to the business – limiting innovation, slowing the pursuit of new opportunities and hindering the shift towards a more data-driven…
AI can write code in seconds. It can turn an idea into a prototype, generate tests, refactor code and help teams move from requirements to working software faster than ever before. As a result, people are questioning the relevance of Agile. Influencers across the tech industry are even declaring that "Agile is dead". We need to ask a different question: What happens to Agile when the ability to build software gets dramatically faster? Agile becomes more important, and more alive, than ever. In its literal sense, agile means the ability to move or think quickly, easily and with flexibility.…
Excitement around data engineering topics is typically limited to data engineers, and their data engineering-adjacent peers. Terms such as “tagging”, “metadata” and “pipelines” are often used without much explanation, which can create gaps in understanding across technical and non-technical teams. That gap can be detrimental to an organization’s technical maturity, and at the very least increases frustration levels across technical and non-technical employees. This tension often bubbles up to funding decisions, when those holding the purse strings are trying to determine which initiatives to…
We are currently suffering from a bizarre industry-wide obsession: prompt engineering. Every week, a new “prompt wizard” posts a cheat sheet claiming that if you frame your request with the right magic incantations “Act as a Senior Principal Architect with 20 years of experience...” the AI will suddenly emit flawless, production-ready software. This is not software engineering. This is spell-casting. The fundamental flaw of the prompt engineering Illusion is simple: natural language is ambiguous, and LLMs (large language models) are non-deterministic black boxes. You can spend hours tweaking…
By Published: September 03, 2026 Long before GenAI agents were used in software engineering, we already had the concept of the '10x Developer'. They weren’t exactly ten times faster, but they did deliver value far more effectively and efficiently than average. But what did it take to create a high-performance team producing high-quality software at a high pace and with a high frequency of change? Enabled by culture and guided by principles, they followed practices that aimed for engineering excellence and superior product quality. It was an environment where design decisions and complex…
For decades, wealth management has operated on a predictable cadence: a dedicated advisor, a quarterly review and the assumption that markets, opportunities and client needs move slowly enough to wait for the next scheduled meeting. In reality, they don’t. This is the story of Marcus, a typical high-net-worth wealth customer who discovered that reality the hard way, and how a new-age wealth management firm’s agentic architecture turned continuous attentiveness into its strongest competitive advantage. The meeting that came too late Marcus is the client every wealth management firm wants: a…
The engine can produce both inferential and deterministic knowledge. A relation the engine surfaces but hasn't confirmed stays inferential, an open judgment call, until a domain expert reviews it. Once approved, the ontology, the relation rules and the ownership record behind each one all become computational, deterministic structures an agent queries through a context-as-a-service layer, a third source of knowledge alongside training data and the vector database. Served this way, guide gives an agent the cross-domain knowledge it needs to get the answer right the first time. Sensor Sensor,…
Related: Evaluating AI agents in production: A practical framework The efficiency paradox: Why speed is not alpha Summarizing a 100-page filing in 30 seconds does not create an alpha-generating edge. As access to the same AI capabilities becomes widespread, the advantage from processing consensus information faster is likely to compress. The differentiator shifts toward proprietary data, better models, better interpretation and better execution. To convert research productivity into true outperformance, institutional frameworks must focus on two higher-order capabilities: Counter-consensus…
An agent failure that looked like success We have a good example of the challenge of agents. A healthcare analytics platform we were working with allowed insurance analysts to query provider performance and claims data using natural language. One clinical reviewer asked the agent: "Show me readmission rates for Medicare Advantage members with CHF (congestive heart failure) diagnoses since January 2024." The agent returned a number that looked reasonable. However, the reviewer was able to flag three mistakes: It included both Medicare Advantage and Original Medicare members. It used an…
An agent is a delegation of human judgment. You authorize it to decide something in a real-world case, what you hand over is not a task but the judgment the task requires. The moment you delegate any judgment, one need follows immediately: the agent must act faithfully. What holds it faithful is a set of mechanisms placed around it. These mechanisms are controllers, terms that have circulated since early 2023 but emerged separately with no shared definitions, leaving loose, overlapping boundaries. This article, from a practitioner's perspective, tries to define and compare them clearly. A…
TL;DR: We post-trained NVIDIA Nemotron 3.5 Lightning for legal and healthcare in a few hours on a single node. The legal adapter was preferred over the base model in 75% of blind comparisons and more than doubled accuracy on a public legal benchmark; the healthcare adapter was preferred in 60%. We fine-tuned away two thirds of the model's recognizable 'AI writing' tells with no measurable loss in writing quality, vocabulary or reasoning benchmarks. The model ships with speculative decoding enabled by default, worth 1.5–2x faster generation out of the box. Always-on agents complete complex,…
Read at the source
Your visit, your choice.
Optional Google Analytics helps us understand visits. Microsoft Clarity records masked interactions to improve the site. Optional tools stay off unless you choose them. Privacy details.