Brad Terrell

LLMs can ace the bar exam—yet still spin fairy-tales when you ask why. Models first solve then invent the reason. A 2024 causal-mediation audit of 12 frontier models showed that in most runs the answer was unaffected by the chain-of-thought the model just wrote. In other words, it solved the problem first and invented a rationale later. Relying on that single rationale is no different from accepting a memo with the signature page missing. High-stakes workflows need stronger guardrails. Here’s the engineering pattern that is quietly becoming the norm across audit-sensitive teams: ⸻ 1 | Retrieval-first, Reasoning-second Ground the model before it thinks. Pull every plausible candidate record from your ERP with deterministic SQL and turn the model’s job into ranking a finite list. Retrieval-Augmented Generation (RAG) has already cut hallucinations by anchoring answers to real documents and citations. 2 | „Four-eyes” architecture Split generation and verification. Agent #1 proposes; Agent #2 interrogates the reasoning, cross-checks retrieved evidence, and either signs off or sends it back. The four-eyes principle has kept banks and insurers honest for decades. Recent research shows multi-agent setups without an explicit verifier still fail up to 60 % of the time—so design that checker in from day one. 3 | Full-stack logging & replay Persist every prompt, retrieval call, ranking decision, and veto. Come audit season you can replay the entire pipeline, not merely wave at a black-box answer. ⸻ Take-home for CIOs/CFOs: Explanations that aren’t causally linked to outcomes are compliance liabilities. The safe path is retrieval-anchored inputs, dual control of reasoning, and traceability that an external auditor can step through click-by-click. (We follow the same pattern inside Hypatos’ agentic automation stack, but the blueprint is open for anyone who values audit-grade AI.)

11

The Agentic Future Demands an Open Semantic Layer. Salesforce is making a first-of-its-kind commitment with an alliance of industry leaders to build towards an open, interoperable future required to accelerate agentic AI. ttps://https://lnkd.in/eFWZic63 Salesforce #agentic #ai

4

Most CROs last 18 months. Chris Degnan lasted 11 years scaling Snowflake from zero customers to $4bn ARR and $100bn+ in market cap. On today's GTMnow Podcast, Chris unpacks how he built one of the most disciplined, high-velocity GTM engines in SaaS history. From selling a stealth product with no customers, to leading a global revenue org through four CEOs and one of the biggest IPOs ever. My 5 takeaways 1️. Be a student of what you sell. Know your product better than anyone. Chris not only mastered but helped create Snowflake’s pricing, contracts, and roadmap, earning credibility with both engineers and customers. 2️. Hire for grit, not pedigree. Startup sales is chaos. He looked for self-starters. People who’ve built from scratch, not inherited playbooks. 3️. Sales and Marketing are one engine. He and CMO Denise Persson measured only one thing: qualified meetings. Not MQLs. Not leads. Alignment or chaos: founders choose. 4️. Customer success is everyone’s job. Early on, he scrapped CS as a standalone team, embedding it into sales and SEs. Accountability for outcomes sat with the people closest to the customer. 5️. Urgency compounds. The best operators act now. Denise Persson’s mantra (“Let’s do it right now”) defined Snowflake’s tempo and culture.

362

38 Comments

Chris Degnan joined Snowflake as employee #13 (and the 1st sales hire). He scaled the sales org from 0 to over $3B in ARR, spanned four CEOs, and retired as CRO after 11 years. In his first podcast post-retirement, Chris opened his CRO playbook, from early enablement to hiring rigor and fending off threats from competitors. If you’re a founder or running sales at a startup, this one is for you… ➡️ Degnan’s non-negotiable rituals for building a world-class sales org #1: Always – Obsess over enablement. It isn’t a nice-to-have—it should be a top priority. Pick a methodology (like MEDDIC) and hold a weekly enablement session with every rep. #2: Weekly – Every single person in your sales org needs to have a forecast call every week. There should also be a direct team meeting. #3: For milestones – Run sales kickoffs & quarterly business reviews. Chris was shocked by how many startups skip these fundamentals. If you don’t invest in enablement and real coaching systems, he says you’re just asking for failure. Don’t try to reinvent the wheel here. ➡️ Should you hire sales reps in less than 2 weeks? If you’re trying to interview dozens of people for one role, the process is set up to fail. What if you were limited to two people and had to decide if either is your person? Much simpler. At Snowflake, the sales team would limit hiring processes to two weeks. Both sides needed to commit to a decision by the end of that window. If a candidate just “wanted to get to know people” and wouldn’t be ready to sign by day 14, they weren’t in the process. This requires a good pitch to candidates and keeps everything efficient. ➡️ The one thing that makes a great sales leader The best measure of a leader is the performance of their individual contributors. Chris believes great sales leaders do three things: they hire great people, they train those people well, and they drive individual productivity. At Snowflake, his ultimate success metric was whether reps could consistently generate $250K in new ARR per quarter by month seven (after they were fully ramped). Full episode: https://lnkd.in/djfzCd7e

362

13 Comments

Big news: Snowflake Ventures is leading a strategic equity financing in AtScale. This investment validates the critical role of semantics in the modern enterprise data stack. As enterprises consolidate data on Snowflake AI Data Cloud, a unified, governed semantic layer is essential for consistent metrics and business logic. A semantic layer like AtScale makes accessing that data more accurate, accessible, and efficient by bridging the gap between business language and data language. AtScale’s Universal Semantic Layer provides this single source of semantic truth for customers like Fidelity Investments and The Home Depot. The company remains independent and interoperable, supporting a broad ecosystem of data platforms, BI tools, and AI applications. This is a good moment to reflect on the future of data and AI. New to semantic layers? I'll post a tutorial link below. And here are five big topics from the 2025 Semantic Layer Summit (link to Substack articles on these below): 1️⃣ Semantic layer standardization is coming 2️⃣ Semantic layers are BI’s Gift to AI 3️⃣ Semantic layers make AI 3X more accurate 4️⃣ Long live SQL (with semantic layers to help) 5️⃣ The great debate: Should semantics be in your database? This is a pivotal moment for the data industry, and I’m thrilled to be a part of it! https://lnkd.in/eVJ276Pp

88

10 Comments

Salesforce announces purchase of Informatica to accelerate its move into an AI-first world 🌍 After talks fell apart for a $10B+ acquisition of Informatica in 2024, Salesforce is now acquiring the company for $8B, a 30% premium to Informatica's current market valuation. Salesforce was early in its realization that owning the data layer was going to be the scaffolding for the entire enterprise software stack, and this acquisition is the next big move in owning that. The real story is how this acquisition fits into their AI operating system strategy, especially around Data Cloud and Agentforce, Salesforce's fastest-growing segment at 120% YoY. Going a bit deeper: For AI agents to work safely and autonomously, you don’t just need more data. You need clean, contextualized data that comes with best-in-class access management and security. Guess where Informatica shines? Enterprise-grade data governance Master data management capabilities Deep pipelines into legacy systems and cloud environments Tools for unifying data across silos In its next act, Salesforce aims to go beyond being a system of record and deploy autonomous agents that act across workflows, including sales, service, and marketing. Doing that with trust and scale requires a real-time, governed, enterprise-wide data layer. There is an argument that this is the final piece of the stack: MuleSoft integrated systems. Tableau visualized them. Data Cloud centralized them. And now, Agentforce will act on them. Informatica will make those actions safe and reliable. While there will be tons of work between now and that vision becoming a reality, you can see a path to Salesforce becoming the first true enterprise AI operating system. Strong move #CRM!

16

AI isn’t theoretical anymore. It’s moving markets. IBM lost roughly 10% of its market cap after Claude demonstrated it could automate parts of a legacy workflow tied to decades of consulting revenue. At the same time, cybersecurity names like CrowdStrike have faced valuation pressure as AI begins automating tasks once considered defensible moats. And in private credit, the ripple effects are showing up fast. Blue Owl’s recent volatility and redemption pressure have put the broader private credit market on alert. When AI can compress margins, change cost structures, or disrupt portfolio companies overnight, that becomes a credit consideration, not just a tech headline. Meanwhile, Anthropic and OpenAI are raising massive funding rounds to accelerate deployment. This is not experimental capital. This is infrastructure-level scaling. But here is what matters for private credit leaders: AI is no longer a strategy conversation. It is an operational and underwriting variable. The firms that will outperform are not the ones talking most about transformation. They are the ones doing three things well: Automating one high-friction workflow at a time Proving measurable ROI in 30 to 60 days Embedding AI directly into existing operational flow No massive platform overhaul. No multi-year roadmap theater. No forcing teams to change everything at once. In this rate environment, with tighter spreads and higher scrutiny, AI can not be abstract. It has to remove work. Reduce risk. Or increase throughput immediately. The market is already repricing companies based on AI exposure. The next step is firms repricing themselves based on AI readiness. Ad Astra ⚡ Ogenticai.com

5

A.I. Clouds are not all the same. Justin Gage of Amplify Partners shares, in explicit detail, how Modal is just built different, and why it's a mistake to think of them as an "inference provider". https://lnkd.in/gmaYQZtT

55

So the new Gartner MQ for CDPs is out. I sometimes wonder if (like that annoying classmate) Gartner deliberately provokes for attention, but in any case, buyers should not rely on their rankings. Rather than a long post, I'll offer a short list of observations: - As usual, G misses big chunks of the market - Note some quite radical shifts versus 2024; no: the market actually didn't move that fast....G is just always five years out of date, and so any given quadrant can't be trusted - "Ability to execute" (Y axis) always correlates with vendor sales and marketing spend - "Vision" (X axis) is PPT theater and fundamental mis-reading of what enterprises look for in this market - Never trust software that Salesforce builds itself... 😉 Link to last year's post on this topic in the first comment....

124

23 Comments

Fascinating new data from a16z on how CIOs want to buy AI apps: 1. CIOs now *prefer* usage-based models. So much for usage-based pricing being seen as a barrier to adoption 🤷‍♂️ My two cents: most AI spend w/ the CIO is on LLMs and compute. Both are usage-based categories where folks have become accustomed to granular pricing and not buying shelfware. With usage-based models, there’s also a connotation that prices will come down over time as vendors compete. Avoid picking a pricing metric that positions your app as a commodity w/o differentiation! 2. Seat-based models aren’t the key to unlocking budget. Only 21% said seat-based pricing was their preferred model. That’s slightly less than hybrid models (23%) even though hybrid models are inherently more complicated with more risk of overages/extra fees. My two cents: buying in a way that reflects the value received is >>> than having perfect budget certainty/predictability. And flexibility matters as folks are finding which AI apps are “experiments” vs ready for “production”. 3. There’s a lot of work to educate CIOs about how to pay for outcomes. 15% said outcome-based models are their preferred way to buy — although that’s still significant considering only *5% of companies* have outcome-based pricing today (based on the 2025 State of B2B Monetization report)! We’re still in the early innings. Among the concerns CIOs raised: (1) lack of clear, measurable outcomes for most products - 47%, (2) unpredictable or unscalabe costs - 36%, (3) difficulty with attribution to a specific tool - 25%. My two cents: I suspect line of business is much more open to outcome-based pricing than the CIO — after all, the CRO and the CMO are measured based on outcomes, too! But if you’re thinking about outcome-based pricing, you need outcome consistency, attribution, measurability and predictability (CAMP). — Cool to see this new data. Very fascinated to see how it changes for next year. #ai #genai #pricing

248

39 Comments

"AI deals convert at 47% vs. traditional SaaS's 25% rate, because the productivity gains are immediate and measurable," so says Menlo Ventures just-released report, 2025: The State of Generative AI in the Enterprise. Unsurprisingly, the report also highlights the fact that GenAI is "the fastest scaling software category in history." A must read. Link to the report in the comments. #FutureOfWork #AI

3

1 Comment

In this article we show how to integrate MinIO to Geyser Data Cloud Archive. A terrific use case for those that having Minio on prem for the high performance AI storage and need a secure, cost efficient place for the cold data! #TaaS #ColdDataCloud #AI https://lnkd.in/g-3qPt9r

44

1 Comment

Today let's talk about Parallel Fan Out/ Concurrent Agentic Design pattern and when parallel agents go rouge What’s the Concurrent Orchestration Pattern? Imagine a scenario where multiple AI agents—each with its own lens or specialty—tackle the same task simultaneously. Instead of a single, step-by-step chain, tasks fan out to different agents in parallel. Then their outputs are merged or aggregated for the final answer. It’s the AI equivalent of a brainstorming session where everyone chips in together. This pattern thrives when you need diverse insights or speed—think ensemble reasoning or reaching a verdict faster. This pattern is commonly used in agentic RAG. Here’s how you can apply it in a Retrieval-Augmented Generation (RAG) system: Step 1: Break the user’s query into smaller sub-queries (e.g., “Define concept X,” “List use cases,” “Give examples”) and map out their dependencies. Step 2: Run those sub-queries in parallel—agents fetch context or process each in isolation (e.g., document retrieval, summarization, external tool usage). Step 3: Once all agents have results, aggregate the findings into a single, coherent response. This technique lets you parallelize independent parts, reducing latency while maintaining clarity—especially effective when sub-queries don’t depend on each other. Super Important : Try to avoid this pattern and use sequential execution even if it's slow, but it might yield a good result, below is the reason As Cognition [the company behind devin] warns in a famous blog post “Don’t Build Multi-Agents”, things can go sideways fast if agents don’t share context or coordination is weak. When agents operate in isolation—making decisions based solely on their own view—the final result might be fragmented, contradictory, or just plain incoherent. Think two agents building different puzzle pieces that don’t fit. The core issues: Context fragmentation: Each agent works in a silo, leading to mismatched assumptions. Implicit decisions: Agents’ outputs reflect unspoken choices that may clash when merged. Coordination complexity: Without strong orchestration, integration becomes error-prone. References : 1. Concurrent Execution Pattern : https://lnkd.in/gZybgm2s 2. LLM Compiler whitepaper : https://lnkd.in/gWwGJhbi 3. Coginition blog Don't build multi agents which shows parallel execution can yield agent drift : https://lnkd.in/gkDKC4TP

14

3 Comments

My mantra for GTM value creation in PE/VC has always been "Focus on the Buyer and the Fundamentals" and "Beware of Shiny Objects". But what happens when buyer behavior changes and the fundamentals also change?! Well that's what is happening with Generative AI. Buyer use of LLMs has changed "top of funnel". Awareness, research and consideration is collapsing into a single LLM conversation that results in a vendor short list. Meanwhile, SEO and SEM driven traffic is shrinking (but it still matter) It was a privilege to moderate a conversation among top PE/VC executives at Pepper's Index 2025 Conference to explore what this all means for our portcos: - Paul Daugherty: AI Advisory Chair TPG and former CTO and Group Chief Executive of Technology at Accenture - Jonathan Metrick: Chief Growth Officer at Sagard - Dev Khare: Partner at Lightspeed - Neal Behrend: SVP Advisory at Insight Partners Key Insights and Takeaways - GEO is Reshaping Discovery: Major drops in web traffic are occurring as users transition from Google and other engines to direct LLM platforms like ChatGPT and Gemini for discovery, comparison, and decision-making - Revenue vs. Traffic Focus: Companies must shift from tracking traffic metrics to measuring real conversion, revenue, and visibility across generative platforms. Leading brands are experimenting with new tools for omnichannel search visibility - Arbitrage and Freshness Windows: Success in the new environment means rapidly testing new strategies such as using proprietary data and frequent citation/refresh cycles, before competitive windows close. Automated pipelines for insights and updates differentiate winners from those left behind - AI as a New UI: AI-driven agents are quickly becoming the primary interface for consumer and B2B interactions. Web presence is less a conversion tool and more a training ground for these bots to shape agentic behavior and decisions - Human + AI for Lasting Impact: The most resilient companies combine sophisticated automation with human expertise, especially in areas where trust, QA, and high-specialization matter. Product alone isn’t enough, process mastery and proprietary insights create real moats Shared Points of View - The End of Traditional Moats: Product-based advantages are fading as barriers to switching drop; workflow sophistication, outcome-based models, and ecosystem integration are the new sources of defensibility - Team Structure and Skillsets: Marketers and leaders must adapt by blending custom AI solutions with scalable human oversight. The future belongs to agile organizations that master both automation and differentiated workflows - GEO as the Next Branding Platform: New brands will emerge, leveraging AI to offer uniquely tailored products and experiences. The panel underscored the opportunity for incumbent enterprises to transform legacy advantages into AI-infused service paradigms It was a great conversation, check out the YouTube link in the comments...

67

9 Comments

Why 95% of Enterprise AI Projects Fail to Deliver ROI A recent report from MIT revealed that nearly 95% of enterprise AI initiatives fail. Unsurprisingly, this has created a wave of concern among investors. While AI has shown promise in areas like content generation, Q&A systems, and workflow automation, most of these applications deliver incremental productivity gains rather than direct, measurable ROI. In many cases, they improve user experience but stop short of driving meaningful revenue impact—unless they are tied to eliminating entire job functions, which is neither practical nor desirable from a human or organizational perspective. This dynamic is not new. We saw something similar during the era of “digital transformation.” I was directly involved in a large-scale banking project where the focus was on enhancing customer journeys, experiences, and backend automation. While valuable, these initiatives often struggled to demonstrate clear revenue impact, which made ROI difficult to justify. AI is experiencing a comparable reality today. To truly move the needle, AI must be deployed in areas that directly influence revenue and cost. Personalized marketing can sharpen messaging and drive sales. Intelligent product and service generation can open entirely new revenue streams. Supply chain optimization can significantly reduce costs and improve efficiency. These are the levers where real ROI will emerge. The challenge is that such initiatives are harder to execute. They demand high-quality data, advanced machine learning, and fine-tuning—capabilities that many enterprises lack today. Yet, with the right talent, vendors, and above all, strong data foundations, AI has the potential to deliver transformative returns. The gap is not in the promise of AI itself, but in how we choose to deploy it.

3

2 Comments

Is your AI stalling?  Here's an 8-question test that explains why. ☐ Do your critical datasets have documented lineage and provenance? ☐ Are your policies enforceable in real-time — not just documented? ☐ Can you deploy a new AI use case in weeks, not months? ☐ Are semantic definitions applied consistently across all your systems? ☐ Can you explain how any AI system reached a specific decision? ☐ Is data preparation automated — or manual and project-by-project? ☐ Is data quality actively monitored with defined SLAs? ☐ Is policy and regulatory compliance evidence automatically generated? If you answered 'no' to more than three of those, your AI initiatives will continue to stall — regardless of how much you spend on models. This is not a technology problem. It's a semantic problem. For 40 years, enterprises made a bet: store the data, lock the meaning in application code, spreadsheets, and tribal knowledge. Seemed reasonable. The interest has been compounding ever since. AI is the margin call on Semantic Debt. AI projects don't fail because the model is weak. They fail because data lacks the meaning and context that machines need. "Garbage in, confidently scaled garbage out." The EU AI Act is enforceable now. Basel IV demands explainable AI. Enterprises will run 35–50 models in production by the end of this year. Manual governance at that scale is not a strategy — it's a liability. 2026 is the inflection point. This gap cannot be crossed incrementally with fragmented tools. It requires a new architectural layer. —— The Semantic Operating System is here. Infinity Data AI's Semantic OS is GA. Our Enterprise Knowledge Model — the core of the Semantic OS — makes enforces meaning: explicitly, machine-interpretably, at the point of every AI interaction. Fast: Data preparation that consumed weeks per project compresses to days. AI deployment cycles that stretched to quarters shrink to weeks. Yours: We activate the knowledge your systems already contain. No rip-and-replace. No long term dependency. You own the model. You own the governance. You own what you build. Proven: Production proven. Zero audit exceptions. Tier-1 regulated bank. Not a roadmap. Not a pilot. Enterprise-grade and available now. —— If this resonates, I'd like to hear where you are. DM me if you are: → "STALLED" — if your AI pilots aren't reaching production → "IN DEBT" — if you know your semantic foundation needs work → "READY" — if you want to see what the Semantic Operating System looks like in your environment No pitch deck on the first call. Just a conversation about what's actually blocking you. The Semantic Debt margin call is here. The foundation for trustworthy AI at scale exists. Let's talk. — Lee Dittmar, Co-Founder, Infinity Data AI 🔗 infinity-data.ai #SemanticAI #EnterpriseAI #AIGovernance #DataReadiness #SemanticDebt #GenerativeAI #EUAIAct #InfinityDataAI

7

1 Comment

Want to hear more from the Spotlight team on the news regarding the sale by Gartner of Capterra, Software Advice, andGetApp to   G2 - this post by John Rockhold is a great summary - and tells you why user reviews continue to grow in importance for LLM’s and by extension, for Influence Orchestration! 

14

Mergers, acquisitions, and divestitures are among the most complex and high-risk activities any organization can undertake. TBM can be a powerful tool to architect strategic and measurable value.

26