David Arnoux’s Post

AI strategy advice: Stay nimble or get stuck holding the bag The AI hype cycle right now feels like watching everyone panic-buy during a flash sale—except the sale never ends, and the prices keep getting better. Here's what's keeping me up at night (in a good way): I'm seeing companies rush to lock themselves into year-long AI contracts because "everyone's doing it" and the FOMO is real. Meanwhile, costs are dropping rapidly, and capabilities are advancing every quarter. The smarter play? Embrace AI aggressively, but keep your options open. For our dev tools as an example, we're going with team-based month-to-month models (shout-out to Claude Code) instead of committing to lengthy enterprise contracts. Not because we're hedging our bets, but because the landscape is moving too fast to get married to any single vendor right now—even the good ones. The other thing nobody wants to say out loud: AI expertise is rapidly becoming table stakes. In 2-5 years, most engineers will be proficient with these tools, whether they're using them for personal productivity or building AI into your systems. We don't need to compete for "AI talent" with Silicon Valley salaries when the skill gap is closing. And here's the reality—most businesses are consumers of AI, not competing to build the next foundation model. That CS degree with an AI specialization is nice to have, but it's not a necessity for using these tools effectively. The takeaway: Be aggressive about adopting AI, but stay flexible on HOW you adopt it. The companies that win won't be the ones who signed the biggest contracts earliest—they'll be the ones who stayed nimble enough to pivot as the technology evolves. What's your approach? Are you locking in or staying flexible?

No shade intended in this post. I stumbled upon a VC post on LinkedIn, that was referencing how DeepSeek hype was over, which is a very intresting perspective. Ironically, DeepSeek just released DeepSeek-OCR two days ago, a model that compresses text 10x by converting it into images. Like what? It can process over 200,000 pages daily on a single GPU, using 61% fewer tokens than competing models while maintaining 97% accuracy. In other words, It's about to get 20x cheaper, and this hits the nail on the difference between hype and raw innovation. Western AI giants are stuck in a ship-fast-or-die mentality. Remember GPT-5 in August? Users hated it: 50% strictly negative sentiment versus only 11% positive. It was marketed as "PhD-level, yet GPT-4 was still our go-to. DeepSeek took the opposite approach. Operating under US chip restrictions, they turned constraints into breakthroughs. Their R1 model was, and still is a mission impossible by industry standards. It is fully open-source under MIT license. No gatekeeping, no marketing circus. Just transparent research that anyone can verify, modify, and build upon. From an AI consulting perspective, DeepSeek stands out for secure enterprise deployment. Chasing every shiny release is tempting, but lasting data security can only exist when the code is open and auditable. Matter of fact, most of my peers in AI & Automation still prefer R1 and 4o models over newer models. New doesn't mean better. Monthly releases don't equal progress. But we get it, Silicon Valleys mantra of “ship fast, fail faster” is apealling. But is it sustainable? And at what cost?

Roadmap to Building Your First AI Agent Creating an AI agent can feel overwhelming , but breaking it down step-by-step makes the journey clear. Here’s a simple flow I follow when designing agents: ➡ Define the Purpose → What problem does your agent solve? Who benefits? ➡ Pick the Right Framework → LangChain, AutoGen, or CrewAI depending on workflow needs. ➡ Select the Model → GPT-4, Claude, or LLaMA 2 based on reasoning, cost, and customization. ➡ Plan Capabilities & Tools → Core functions, integrations with APIs, databases, productivity apps. ➡ Design the Architecture → Input handling, reasoning, memory, and error management. ➡ Implement Memory & Context → Short/long-term recall, embeddings, personalization. ➡ Enable Safety, Monitoring & Speed → Filters, analytics, optimization. ➡ Iterate & Expand → Continuous learning, multimodal capabilities, personalization, deployment, and upgrades. This roadmap helps transform an abstract “agent idea” into a reliable, production-ready system. 💡 I come from a fintech and payments background, so I’ve seen firsthand how these steps apply to secure, compliant, and scalable AI systems. Whether it’s RAG-powered assistants for financial insights or workflow-driven payment bots, the process remains the same.

🚨 The AI infrastructure war just got a lot more interesting. CoreWeave's acquisition of Monolith AI isn't just another tech deal—it's a chess move that could reshape the entire enterprise AI landscape. Here's why this matters: While everyone's been focused on ChatGPT and consumer AI, the real money is in industrial applications. Manufacturing, engineering simulations, complex modeling—that's where AI delivers massive ROI. CoreWeave just positioned themselves to own that space. Monolith AI brings cutting-edge simulation technology that turns CoreWeave from "just another cloud provider" into a specialized AI powerhouse for enterprise workloads. The timing is perfect: ✅ Companies are desperate for AI solutions that actually move the needle on their bottom line ✅ Industrial AI requires massive computational power (CoreWeave's specialty) ✅ Most cloud giants are still playing catch-up in specialized AI infrastructure This acquisition signals something bigger: we're moving from the "AI hype phase" to the "AI implementation phase." The companies that win won't be the ones with the flashiest demos—they'll be the ones solving real industrial problems at scale. CoreWeave just made their move. Who's next? What's your take—are we finally seeing AI move from buzzword to business-critical infrastructure?

Remember that knot in your stomach? The one that tightens when you hear "AI will replace everything"? For a while, even in the low-code/no-code world, there were whispers. Would all the automations I build with n8n and Make become obsolete? Would my AI agents just take over entirely, leaving no room for human ingenuity? I've spent years diving deep, sometimes failing, sometimes celebrating breakthroughs, building everything from intelligent CRMs to hyper-efficient chatbots. My journey with tools like n8n, Make, and crafting AI agents has shown me something profound: AI isn't a threat; it's the ultimate co-pilot. A recent report just confirmed what I've been experiencing firsthand. AI isn't here to end low-code/no-code; it's here to supercharge it. It's democratizing development, letting non-developers create alongside seasoned pros. It's optimizing, debugging, and automating the very process of building. Think about it: smarter automations, more adaptable applications, and a massive reduction in barriers for anyone to experiment with AI and Machine Learning. This isn't just theory. We're talking about hyperautomation becoming a reality. Projections say that by 2025, 70% of new enterprise applications will use low-code or no-code technologies. AI is the engine driving that acceleration, making solutions I build with Flowise, Typebot, and various AI APIs more intelligent and responsive than ever. So, what's your take? Are you seeing AI as a catalyst or still feeling that knot of fear? Share your thoughts below, or if you're ready to explore how AI can amplify your low-code/no-code projects, send me a DM. Let's build something smarter together.

Ever Wonder Why Sovereign AI Matters? Take a closer look at this chart — it’s a snapshot of today’s AI economy. The same few companies owning the infrastructure, the models, the chips, and increasingly, your data. We’re told there’s choice — that innovation is open and democratised. But when the same five or six players control the pipelines of intelligence, what we’re seeing isn’t diversity — it’s centralisation disguised as progress. The reality? Many organisations are willingly giving away their IP and proprietary data in the rush to integrate “AI” as fast as possible. For the illusion of innovation, they’re trading their most valuable asset — information — to systems they don’t control, on infrastructure they don’t own. At Nevari, we believe this is the critical moment for Sovereign AI — AI built privately, ethically, and strategically within your own enterprise ecosystem. Sovereign doesn’t mean isolated — it means secure, compliant, and owned. Data is worth more than gold. More than oil. It fuels your products, your people, your innovation — and your competitive advantage. Giving it away for the hype is like signing away the blueprint to your own future. There are safer, smarter, and more secure ways forward — open-source models, private LLM architectures, and trusted frameworks that let you innovate without surrendering control. Nevari’s mission is to help organisations navigate this intelligently: • Build AI-first infrastructures without dependency. • Embed data governance and IP protection at every level. • Create measurable ROI — not marketing hype. Because the future of AI isn’t about who has the biggest model — it’s about who owns their intelligence. That’s why at Nevari, we say it clearly: Sovereign AI or nothing.

AI Won’t Make You Rich — But It Might Make Everyone Else Smarter Jerry Neumann’s essay “AI Will Not Make You Rich” is one of the most interesting pieces about AI I’ve come across recently. When shipping containers debuted in the 1950s, they completely changed global trade. Costs collapsed, efficiency exploded - but, unlike the personal‑computer revolution that followed decades later, very few people actually got rich from it. Competition, capital intensity, and fast adoption made it almost impossible for founders or investors to build defensible positions. The big winners turned out to be companies that used the technology well - names like IKEA, Walmart, and Costco. Neumann argues that generative AI is now in a similar spot. It’s a breakthrough technology, but one that’s arriving at a mature stage of the market - competition is already brutal, and the biggest players own most of the infrastructure. Large model builders will dominate because of scale and capital intensity. Smaller challengers will have a hard time standing out, and app‑focused startups will find their ideas copied, priced down, or rolled into larger platforms. Those expecting AI to create fortunes like the microprocessor wave may find results closer to containerization - massive impact, modest profits. The real opportunity sits further downstream. Sectors such as professional services, healthcare, education, finance, and creative industries stand to gain the most - if they can execute well. The technology will help them cut costs and increase productivity, but success will come down to how they run their businesses day‑to‑day. Most of AI’s benefits, though, will flow to consumers - just as previous waves of automation did - through better, cheaper, and more accessible products and services. For investors and entrepreneurs, the key isn’t chasing “the next big thing.” It’s finding and backing the companies that actually use AI to open new markets or run more efficiently. Value builds quietly - through good execution, smart adoption, and focused differentiation. (Link in comments)

Most mid-sized companies don’t need “AI strategy.” They need to identify a few areas where it seems like it could be helpful and get some quick wins. The places where we’ve been able to roll AI out the door quickly for our mid-sized customers have been in rapid prototyping and in accessing untapped internal data. Rapid prototyping. We’re big believers in Lean Innovation. In fact, the company name Lineate is meant to invoke improving understanding through iteration. Using AI well can accelerate this process dramatically, getting from idea to truly functional prototypes in days, not months. It enables product teams to really experiment in real time as the application evolves. Data access. Pulling structured, unstructured, and behavioral data from across the org and making it usable. Especially for data that is buried in documents or log files, or more complex behavioral data that is just available in corporate databases. This kind of thing is pretty central to AdTech and useful in some of the FinTech and other work we do, but a lot of companies just didn’t have the bandwidth to invest in accessing this kind of data when it’s not core to their business. But AI is changing that, and a lot of stuff that in the past would have taken substantial efforts can now be accessed pretty quickly. Forget “moonshot” AI projects. The ROI is in speed — shipping faster and making sense of data you already have.

💡 For anyone wondering where AI is heading next. Sequoia Capital recently shared their investment thesis for the next 12–18 months. Here are the 5 areas they believe hold the most opportunity: 🔹 Persistent Memory - Long-term context retention so AI can “remember” - Persistence of identity & personality across sessions - Unlocks deeper organizational understanding & wider work adoption 🔹 Seamless AI-to-AI Communication - Beyond MCP: new protocols so AIs talk to each other - Infrastructure layer opportunity - Could erode moats of current “one-click” providers 🔹 AI Voice - Real-time, low-latency conversations now viable (Ribbon ✌) - Consumer apps: AI friends, companions, therapists - Enterprise: logistics orchestration, trading desk automation 🔹 AI Security - Security needed at every layer of the stack - From foundation models to end users - Future may mean hundreds of AI security agents per person/agent 🔹 Open Source AI - Once seen as competitive with state-of-the-art models, now more precarious - Critical to keep AI open and accessible to all builders - Prevents a future dominated by a handful of well-funded giant Food for thought 😁