kengdevLet’s talkJournal.
AI agents, software development and automation: methods, technical decisions and practical experience.
AI ROI: quantify an agent's return before you sign
AI ROI: how to quantify what an agent gives back in an SME, from time actually freed to supervision cost, and the cases where the maths says no.
Read article
Codex vs Claude Code: pricing, limits and guardrails
Codex vs Claude Code on what can actually be checked: both pricing ladders, what each vendor does and does not publish about quotas, and the default guardrails.
Read article
AI agent for business: what actually runs
AI agent for business: the four decisions that determine whether one survives its first quarter, and the three cases where an agent is the wrong tool.
Read article
Open source AI agents: read the licence before the README
Open source AI agents: why half of them are not, in the strict sense, what the licence really allows, and the real cost, which has nothing to do with it.
Read article
Hermes Agent skills: write, install, supervise
Hermes Agent skills: the SKILL.md format, the eight install sources, the four trust levels, and the five things that break in production.
Read article
Hermes Agent on Docker: install it and keep it
Hermes Agent on Docker or elsewhere: the seven runtimes available, what installing really does, and the four things that break once the agent is live.
Read article
MCP server: which one to wire, and on what terms
MCP server: the two ways to run one, where to find the ones that already exist, and the five questions worth asking before wiring one to your own data.
Read article
Claude Opus 5: effort, long agents, what SMEs should change
Claude Opus 5 (24 Jul 2026): same price as Opus 4.8, effort as the main control and a 1M context. What it changes for SME production agents.
Read article
AI agent for SMEs: setup budget, run cost and traps
What an AI agent costs for an SME: setup and run ranges, the real cost of HITL, and what to scope before signing a quote.
Read article
Buzz Jack Dorsey: human + AI agent workspace (buzz.xyz)
Buzz Jack Dorsey: who built Block's open-source workspace, how to install it (desktop, self-host, VPS), what agents can really do, and the SME guardrails.
Read article
MCP 2026-07: OAuth, stateless core, team migration
MCP OAuth and the 2026-07-28 spec: the core goes stateless and auth gets hardened. The SME checklist to migrate without breaking your servers.
Read article
LLMOps: run LLMs in production without chaos
LLMOps: how to run LLMs in production with evals, versioned prompts, observability, token budgets and guardrails an SME can keep.
Read article
Claude Code vs Codex: controls and the token bill
Claude Code vs Codex in July 2026: what moved on approvals, MCP and artifacts, and the team policy that keeps the token bill under control.
Read article
Fine-tuning an LLM or RAG: how to choose
Fine-tuning an LLM or RAG: the grid to decide on your data, budget and risk, plus the traps that cost small companies the most.
Read article
RAG LLM: ground a model on your data without fine-tuning
Production RAG LLM: an SME method to ground a model on your data, from corpus to citations, with eval and cost under control.
Read article
Claude Code: context, compact, clear and session hygiene
Claude Code in depth: manage context with compact and clear, avoid the dump zone, and govern skills, MCP and permissions as a team.
Read article
Kimi K3: open weights, cost and risks for an SME
Kimi K3 (Moonshot, July 2026): a ~2.8T-parameter open-weights model with day-0 vLLM serving. What it changes for SMEs, and how to scope the POC.
Read article
GPT-5.6 Sol, Terra, Luna: what changes for an SME
GPT-5.6 (9 July 2026): the Sol, Terra and Luna tiers, ChatGPT access, API pricing and ultra mode. The SME plan to test without migrating everything.
Read article
Context engineering: making production agents more reliable
Context engineering for business agents: the Write, Select, Compress and Isolate strategies to control what enters the model, tokens and quality.
Read article
AI code review: speed up without automatic merge
AI code review: a fast first pass in comments only, no automatic merge, and a two-sprint noise calibration before scaling it.
Read article
Vibe coding: when to explore and when to specify
Vibe coding: prototype fast without reviewing code, useful in spikes but risky in production. When to switch to specs and RPI once a customer enters.
Read article
AI native: process, lucidity and team delivery
AI-native team: conventions before tools, delivery KPIs instead of commit counts, and a 90-day roadmap to make the shift without chaos.
Read article
Local AI: Ollama, self-host and real selection criteria
Local AI or cloud: a decision grid based on data, TCO, quality and SLA. Ollama for dev, another stack for multi-user production.
Read article
Multi-agent system: patterns when one agent is not enough
Multi-agent systems: router, specialist, critic and orchestrator patterns, plus the Isolate lever, for when one agent is no longer enough.
Read article
Model Context Protocol: wire an agent to your tools without chaos
Model Context Protocol: the Anthropic open standard to wire an agent to your tools, covering allowlists, security and schema token cost.
Read article
AI automation: from isolated chat to pipeline
Enterprise AI automation: the four stages from isolated chat to CI/CD pipelines, with the ROI, governance and guardrails to set.
Read article
Human in the loop: designing supervision for AI agents
Human in the loop (HITL): how to design production supervision for an agent, from thresholds to validation queues, without review fatigue.
Read article
How to create an AI agent in an SME (without picking the wrong target)
How to create an AI agent that earns its keep in an SME: process first, then the LLM, tools and control trio, a first flow that pays, measured HITL.
Read article