Agentic AILLMTrends
AI predictions for the next 3–5 years, from a studio that builds it
Our grounded AI predictions for the next few years, from a team that ships agents and automation, with what each one means for your business.
7 min readSYF Tech
Notes on AI, shipped for real.
Practical notes from building agents, LLM features and automation for healthcare, e-commerce and agriculture.
More notes
12 articles
Agentic AIEngineeringHow an AI agent actually works: the loop, the tools, the guardrails
An AI agent is a language model in a loop that calls tools, reads the results and decides again. Here is how the loop, tools and guardrails fit together.
6 min read
LLMEngineeringHow an LLM produces an answer, and what that means in production
Tokens, next-token prediction, the context window and sampling explain most LLM behaviour in production, from the bill to the hallucinations.
5 min read
Next.jsEngineeringWhat's new in Next.js 16, and why use Next.js instead of plain React
Next.js 16 makes caching opt-in with Cache Components, ships Turbopack by default and renames middleware to proxy. What changed, and when plain React is enough.
9 min read
ReactEngineeringHow useCallback works in React, and why your memo child still re-renders
useCallback keeps the same function between renders until a dependency changes. Why that matters for memo children and Effects, and how stale closures sneak in.
9 min read
ReactEngineeringHow useMemo works in React, and when it doesn't help
useMemo keeps the result of a calculation between renders and only recalculates when a dependency changes. Here is what React actually compares, the three cases where it pays off, the bug that quietly disables it, and how the React Compiler changes the picture.
8 min read
ReactEngineeringuseMemo and useCallback: when to use them, and when not to
A decision guide for useMemo and useCallback. Four questions tell you when memoizing pays off, when to skip it, and what changes with React Compiler on.
9 min read
LLMEngineeringRAG vs CAG: when to retrieve and when to just load the context
Retrieval-augmented generation fetches the right passages for every question. Cache-augmented generation loads the whole knowledge base once and reuses it. Here is how we choose between them.
3 min read
Agentic AIEngineeringAgent or workflow? How we decide before writing any code
Most "AI agent" projects work better as a plain workflow with one or two model calls. Here is the checklist we use to tell the difference.
3 min read
LLMProductivityHow we use Claude every day to get more done
Not a demo, a routine. The habits, setups and guardrails our team uses with Claude for writing, research, code and reviews, and where we deliberately keep a person in charge.
3 min read
HealthcareLLMBuilding AI for healthcare without putting patient data at risk
Language models can take real paperwork off clinicians' plates. These are the design rules we follow so that patient data stays where it belongs.
2 min read
E-commerceLLMWhy your store's search box loses sales, and how AI fixes it
Shoppers type descriptions, typos and mixed languages. Keyword search was never built for that. Hybrid search is.
2 min read
AgricultureAutomationWhat AI can (and can't yet) do in a greenhouse
Sensors and dashboards come first. A practical look at where data and AI actually help growers, and where a good alert beats a clever model.
2 min read
Browse by topic
What are you working on?
Have a project in mind?
Tell us what you're building. A senior engineer replies within 24 hours.