Defaulted to the priciest model for everything. Bill: $25→$500/mo. Same tasks. Routing is the skill now, not picking the smartest model. Sonnet 5 = daily driver. Opus 4.8 = judgment call. Fable 5 = hard problems only. → open.substack.com/pub/creato… #AIProductivity #BuildingWithAI

#AIAgents #ContextWindow #AgenticAI #PracticalAI #AIForwardEngineer #LLMOps #BuildingWithAI #ProductionAI #ArtificialIntelligence #AIEngineering #Anthropic #Claude #AIArchitecture #MachineLearning #AIMode

Anthropic just gave Claude a million-token context window. Most engineers will treat that as permission to load everything in. #AIAgents #ContextWindow #AgenticAI #PracticalAI #AIForwardEngineer #LLMOps #BuildingWithAI #ProductionAI #ArtificialIntelligence #AIEngineering

Check out my latest article: The Agent that could not represent itself linkedin.com/pulse/agent-cou… via @LinkedIn #AgenticAI #SoftwareEngineering #AIStrategy #CriticalThinking #BuildingWithAI #TechLeadership #EngineeringMindset #AIEngineering #ProductThinking #ResponsibleAI

Fable 5 offline 18 days. June 30: back with Sonnet 5 ($2/$10/M) jailbreak severity framework co-built with Amazon, Google, Microsoft. Same week: Claude Science launched. OpenAI offered the US gov a $42.6B equity stake. → thecreatorsai.com/p/fable-5-… #AIAgents #BuildingWithAI

Never ever ever build an LLM app without KV cache. You're paying for O(n²) attention. On every. single. token. Here's what's actually happening under the hood: Transformer attention computes Q, K, V matrices for every token in your sequence. Without cache, generating token n means recomputing K and V for all n-1 previous tokens. For a 4,000-token prompt generating 500 tokens: → 500 full attention passes over 4,000 tokens → Each pass = O(n²) in memory, O(n) in compute per layer → Across 32 attention heads × 96 layers (GPT-4 scale) you feel this KV cache breaks the cycle. Keys and Values are computed once per token, then stored. Only the Query is computed fresh for each new token. Memory cost: O(n × d_model × num_layers) Compute cost per new token: O(1) instead of O(n) Trade-off: At 128k context with FP16, KV cache alone can consume 40–80GB of VRAM. That's why H100s with 80GB fill up faster than you expect. That's why paged attention (vLLM), sliding window attention, and prefix caching exist they're all managing KV cache memory pressure. If your inference latency is spiking at long contexts, check your KV cache eviction policy before you blame the model. Save this before your next AI Engineer Interview. 🔖 #LLMInference #AIEngineering #GenerativeAI #MLOps #KVCache #AIInfrastructure #BuildingWithAI

We built AI agents for local deployment and discovered that hosting them on personal computers enhances privacy and security, Operating rule: prioritize data sovereignty when designing AI solutions. #BuildingWithAI #AIAgents xgenplus.com

That's a wrap on the Kenya AI Challenge 2026 hackathon final. Two days, 100 builders, 8 awards, and solutions built for real farmers facing real challenges in advisory and finance. #KenyaAIChallenge2026 #AIKenya #Neo4j #BuildingWithAI #AgriTech

My take on @figma Config 2026. 👇 ibuildwith.ai/blog/at-figma-… #Figma #AI #Design #BuildingWithAI

Agents are only as capable as their infrastructure. We built it.17 APIs. 125 endpoints. MCP. CLI. Zero Trust. Burst limits. Correlation IDs.Agent-native from day one. datpaq.com #APIFirst #AgentNative #APIInfrastructure #AIAgents #BuildingWithAI #DeveloperTools

Teams pitch their AI solutions for Kenyan smallholder farmers, competing for a share of the KES 300K prize pool. @MC_AgriFin @Lovable @DirectEdDev @neo4j #KenyaAIChallenge2026 #AIKenya #BuildingWithAI #AgriTech

Kenya AI challenge Hackathon Day 1 is underway at @ALXKenya, with teams efining AI solutions for Kenyan smallholder farmers. @MC_AgriFin @Lovable @DirectEdDev @neo4j #KenyaAIChallenge2026 #AIKenya #Neo4j #BuildingWithAI #AgriTech

🚨 The fastest-growing repos on GitHub this June: 1. pewdiepie-archdaemon/odysseus PewDiePie (111M YouTube subscribers) shipped a self-hosted AI workspace and put it up for free. 75.8k stars in under three weeks. 2. mattpocock/skills Matt Pocock's actual Claude Code skills. The TypeScript wizard made his personal .claude directory public. 3. chopratejas/headroom A Netflix engineer's fix for bloated AI context. Compresses what the model reads by 60-95% without changing the output. 4. DietrichGebert/ponytail Trains your AI agent to think like the most efficient dev in the room. Skip the code that doesn't need to exist. 5. calesthio/OpenMontage Production-ready agentic video. 12 pipelines, 52 tools, 500 agent skills in one system. 6. jamiepine/voicebox Self-hosted AI voice studio. Clone voices, generate audio, dictate. No API bill, no subscription. 7. ZhuLinsen/daily_stock_analysis Stock analysis across US, China, and Hong Kong markets. LLM-powered, real-time news, dashboard included. Runs free on a cron. 8. mvanhorn/last30days-skill Pulls from Reddit, X, YouTube, HN, and Polymarket and turns it into one coherent summary. Context included, not just links. 9. bytedance/deer-flow ByteDance's open-source agent for long research and coding tasks. Runs for hours without you babysitting it. 10. DeusData/codebase-memory-mcp Builds a live knowledge graph of your codebase for Claude Code, Cursor, and Codex. Local-only, updates automatically. GitHub in June 2026 is genuinely something. Bookmark this. #GitHub #OpenSource #AITools #BuildingWithAI

Designing with Ai #buildingwithai

Just got accepted into the ATF AI Challenge — Africa's largest hands-on AI program backed by Google.org Python pre-req starts today. AI School kicks off in July. Let's build.#ATFAIChallenge #AfricanTech #BuildingWithAI 🚀

Insurance website ready to use template with its own admin to update content. #buildingwithAI #ProductDesign

Mastering graph databases like Neo4j opens up new possibilities in AI implementation, especially in Agri-Fintech solutions. Explore more at kenyaaichallenge.com/ #KenyaAIChallenge #AgriTech #BuildingWithAI

A computer science student, Del Siata, shares his appreciation for graphs in RAG architecture and his drive to build practical solutions for Africa. Learn more and follow for more: kenyaaichallenge.com/#Neo4j #KenyaAIChallenge #AgriTech #BuildingWithAI

Most people think AI is just a tool for big tech companies and developers. Solvana AI is here to change that narrative. We're building AI that actually works for everyday people simple, smart, and built around real problems that need real solutions. The future of AI isn't complicated. It just needs the right hands shaping it. At @solvana_ai we're not just talking about innovation we're moving with it. Every product, every feature, every decision is rooted in one question: does this actually help people? That's the standard we hold ourselves to. And we're just getting started. Follow along because what's coming next is worth watching. #SolvanaAI #ArtificialIntelligence #TechForGood #AIInnovation #FutureOfAI #BuildingWithAI @solvana_ai

Laura Shavia, a CS student & software engineer, shares her positive experience with the Neo4j pre-hackathon session. Her advice: attend sessions, network, and seize this opportunity! Join the challenge at kenyaaichallenge.com/ #Neo4j #KenyaAIChallenge #AgriTech #BuildingWithAI