Matthew Brandt

Agentic AI is becoming more and more important, but its value and direct benefits are also invisible, even though it could work as an autonomic system for the users. This is the reason why I am still struggling in the "last mile" to learn and apply Agentic AI for business purposes of Regrowth and Restartup. Here is a link helping me to understand such a true reality: 智能体AI如何为现实世界影响奠定基础 https://lnkd.in/d9Ww5p27 智能体AI越来越被设���为不可见的基础设施，在幕后运行，而不是用户直接交互的东西。 1. 智能体AI越来越呈现为系统级转变而非独立功能。 1) 它越来越被定义为不可见的基础设施，而不是用户直接交互的东西。 2) 我们使用的每个应用程序的接口都将是AI，都将是智能体。 3) 智能正被嵌入到在幕后运行的平台、工作流程和服务中，这使得其价值更难以展示——即使其重要性在不断增长。 2. 智能体AI强调自主决策制定、协调和执行能力，它被定位为未来应用的接口层，嵌入到平台和基础设施中，而不是作为单一产品提供。 1) 智能体AI强调了向系统级思维的转变，其中硬件、软件和模型是不可分割的。 2) 这种集成使大规模自主性成为可能，即使其好处不是立即可见的。 3) 它代表着系统级转变，将成为我们使用的每个应用程序的接口。 --- Obviously, is it meaning the product should not be the Agentic AI itself - supposed to hide in an autonomic system? It's so challenging and also enjoyable in a citizen developer way to vibe code such an autonomic system. How to unlock the hidden value of an autonomic system is new as well nowadays. It's really applicable to learn what is explained from the link... I will continue to proceed a massive scale of brainstorming or prototyping anyway - aiming for a big breakthrough! So, a lot of the work should be done in recent peer programming with Lovable AI and MeDo AI. (Both are much better for vibe debugging and vibe fixing as my preferable tools of Collaborating Coding - able to deliver at the end!) 👇By the way, the team of MeDo AI seems younger than Lovable AI team: https://lnkd.in/de38CXBJ 👇🙏👇

5

Agency or in-house team. It's one of the decisions that costs companies the most money when they make it for the wrong reasons. Most decide based on price. That's where the problem starts. An agency makes sense when: → The company needs expertise across multiple channels and can't hire a specialist for each one. → Ad spend doesn't yet justify a full team. → Speed matters: launching campaigns in weeks, not after months of hiring. → An outside perspective is needed from people who've seen dozens of accounts and know what works. An in-house team makes sense when: → Marketing is core to the business model, not a support function. → The product is complex and requires deep knowledge that takes months to build. → Sales, CRM, and operations data need to connect with campaigns every single day. → The brand needs to produce content constantly, with internal context. But what the market is increasingly moving toward is neither. It's the hybrid model. An in-house leader who understands the business, sets the strategy, and owns the data. And external specialists for tactical execution, audits, or specific channels. Artificial intelligence (AI) accelerated this shift. Work that used to take three people can now be handled by one with the right tools. That makes small, strategic in-house teams far more viable. The most expensive mistake isn't choosing an agency or an in-house team. It's hiring execution when what was missing was strategy. Or hiring strategy when no one inside the company can execute it.

1

Analytics Report Service "We're not making data-driven decisions" is code for "we don't have time to look at the data." 📊 I build automated monthly analytics reports from GA4 and your business data — plain-language insights, trend analysis, and prioritised action points — delivered automatically so you actually read them. No more logging into dashboards you never open. From R2,500/month. R3,000 setup fee. First report in 7 days. Drop "DATA" in the comments and let's talk about your reporting. #Analytics #DataDriven #BusinessIntelligence #SouthAfrica #MahloAIConsulting

1

2 Kommentare

Very Interesting summary of AI systems and AI literacy in few words: - LLM writes - RAG knows - Agent acts - Agentic AI manages

12

LLMs and Analytics Everyone's excited about LLMS for writing content. But when you try to apply them to analytics, the challenge is real. LLMs are probabilistic by design. This works great for creative tasks where variation is ok or even wanted. But analytics requires deterministic results. When someone asks "how many units did we sell last quarter?", there's only one correct answer. However, two users can ask the same analytical question and receive different answers. Not because the LLM is broken, but because business questions are inherently ambiguous and LLMs will resolve that ambiguity differently each time. Example from a non-technical user: "What product sold more last month?" There are at least a dozen valid interpretations: "Product" = SKU? Product line? Brand? Category? "Sold more" = Revenue? Units? Profit? Growth rate? "Last month" = Calendar month? Last 30 days? Fiscal period? An analyst knows to ask clarifying questions. A non-technical user doesn't realise they're being ambiguous. And an LLM will happily pick an interpretation, potentially a different one each time. This inconsistency erodes trust. The DATA/AI industry talk about "text-to-SQL" as if it's a solved problem just like that. It's not. Here's what's actually required: Semantic Standardisation: Mapping business language to technical reality. When users say "product," what do they actually mean in our context? Curated Question Libraries: Pre-built patterns for common queries that act as interpretive anchors and ensure consistent interpretation. Layered Context: The LLM needs organisational context (how your business operates), domain context (rules for specific verticals), and data context (what tables contain and how they relate). Progressive Complexity: Layering is paramount with simple counts and sums. Then, introducing ratios and trends. Each layer builds trust and teaches better patterns. This means that we have: LLMs for writing: Embrace variation, encourage creativity. LLMs for analytics: Constrain interpretation, standardise assumptions, validate relentlessly. The technology is advancing rapidly, but the core issue remains: Business language is ambiguous, and analytics demands precision. We need to invest in the unglamorous work of mapping business vocabulary to data reality, building interpretive frameworks, and designing systems that build trust through reliability. #ai #analytics #metadata

77

6 Kommentare

𝐓𝐡𝐞 𝐛𝐢𝐠𝐠𝐞𝐬𝐭 𝐩𝐫𝐨𝐛𝐥𝐞𝐦 𝐰𝐢𝐭𝐡 𝐲𝐨𝐮𝐫 𝐀𝐈 𝐦𝐚𝐫𝐭𝐞𝐜𝐡 𝐬𝐭𝐚𝐜𝐤? 𝐈𝐭’𝐬 𝐩𝐫𝐞𝐭𝐞𝐧𝐝𝐢𝐧𝐠 𝐭𝐨 𝐛𝐞 𝐮𝐧𝐢𝐟𝐢𝐞𝐝. It looks integrated in slides. It behaves like silos in reality. In a sharp breakdown, Tanuj J., CEO of Eulerity, exposes how disconnected tools create the illusion of intelligence while quietly dragging performance down. Because if your systems can’t see each other, neither can your strategy. ✅ 𝐓𝐎𝐏 𝐓𝐀𝐊𝐄𝐀𝐖𝐀𝐘𝐒: 1️⃣ Intelligence needs shared data, not isolated tools 2️⃣ Each platform optimizes a slice, not the whole 3️⃣ Debugging performance becomes a multi-vendor blame game 4️⃣ AI without integration is just automation at scale 5️⃣ Real systems act across channels, not within them 🔗 If your martech stack isn't delivering optimal results, give this a read: https://lnkd.in/dy24XFPh #MarketingTechnology #Martech #Marketing #AI #Data #B2BMarketing #MarketingLeadership #TechIntelPro

5

1 Kommentar