🦀 Rust & Move Development Services - High-performance backend APIs (Actix/Axum) - Smart contracts on Sui & Aptos - Cloud integration (AWS / GCP) - Fractional CTO guidance for scaling startups 🔗 Let’s build something fast, safe, and future-proof. 💬 DM me or visit

mcp tools with clean architecture. supabase direct access. parallel async operations. the enterprises winning at ai arent buying more chatgpt seats. theyre building intelligent infrastructure. whats your ai actually connected to?

here is the part nobody talks about: ai isnt magic. its infrastructure. treat it like you treat your ci/cd pipeline. observable. testable. composable. we open sourced the patterns.

our system_health tool replaced 3 api calls with 1. 300 tokens instead of 800. multiply that by 10,000 daily queries. thats real money.

we built an mcp server that gives claude direct database queries, health checks, and admin operations. one tool call. not four. the old way: ai asks you to run commands, copy paste results, wait for analysis. the new way: ai queries your postgres, checks circuit breakers,

enterprises are spending millions on ai integrations. most of them are building the wrong thing. 🧵 here is what nobody tells you about production ai tooling: your llm doesnt need more prompts. it needs direct access to your systems. #ai #enterprise #mcp #infrastructure

the enterprises asking "how do we use ai" are already behind. the question is "how do we build systems ai can actually use well" your debug tools optimized for humans? probably terrible for ai. fix that first.

we documented the patterns: composite tools that batch operations. summary modes before detail modes. smart defaults with escape hatches.

here is the truth: most ai implementations fail because engineers treat context like its free. its not. every token is latency. every token is money. every token is cognitive load for the model.

the results: 83% reduction in skill file size. 95% reduction in database queries. same insights. fraction of the cost.

ai doesnt need all your data. it needs the right data at the right time. we built a tiered investigation system. scan_anomalies: 500 tokens. quick health check. only if issues found: load the full trace. progressive disclosure for llms.

we had a debug tool returning 1mb of json. thats 300k tokens. for one query. the ai would choke. lose context. forget what it was doing. here is what we learned:

your ai assistant is wasting 95% of your context window. I watched it happen in production. 🧵 #ai #devtools #engineering #enterprise

I tell all my friends to switch to Claude. Friends don’t let friends use openai.

I analyzed insights from a CEO coach, a YC founder, and experts in executive positioning to uncover four unexpected secrets of strategic influence. Strategic thinking is not an inherent magic but a set of learnable skills. Here's what I discovered: 1. **The Riskiest Path is the

I've mostly spend my tokens on Sonnet 4.5 and I am still not convinced to switch to Opus due to the many of the reasons you brought up here, these are the key improvements that I made to help claude cut through the fluff: - provided summarised docs for the tech stack - optimised

Before coding, follow these AI-driven interview steps: - Obtain the Claude Code. - Set it up via your terminal (e.g., Command Line, PowerShell, WSL). - Create a simple project spec file like ` https://t.co/73NSL7s0fc` for `Accounting software for YouTube creators`. - Run Claude

open source highlights include all frame ( https://t.co/eaqM3Kb0cz) a rust ai first framework, allsource core ( https://t.co/jLP6P5xVnN) an event store still being refined, monorepo meta ( https://t.co/9fKSD73HmV) for orchestration, and a bootstrapping repo ( https://t.co/EOiwTRZBbZ)

Ended at iProov, focusing on Rust and TypeScript for AI-era security. The crypto project nears MVP. Claude Code is now central, with Cursor for specifics. Lesson learned: Paths aren't always straight. Stepping back can propel you forward. What did 2025 teach you? #CareerGrowth

Post-Ebury, job hunting in London was tough. Long leadership interviews, AI and remote work focus added complexity. I built: boosted a crypto project, open-sourced my learnings, crafted a Rust AI framework, event store, monorepo tool, and bootstrap repo with MCP.

Rejoined management at Ebury, learned from unexpected financial onboarding challenges. Improved MundoWallet into a Visa-backed multi-blockchain app, with Avalanche pivotal. My AI toolkit expanded from Windsurf to Google Suite, Copilot, Cursor, and Claude Code.