Tech Trends — The Monthly Edit

✳ THE BIG PICTURE AI & Data

The next AI skill? Knowing when it works.

Building an AI demo is one thing. Knowing it works reliably is another. This month, explore evaluation, stronger foundations, and the skills that make products useful.

Explore the skill Source: Anthropic · Demystifying evals for AI agents

FROM “IT WORKS” TO “IT HOLDS UP” ↗

01 Define the task✳02 Test the result

EVALUATION CHECKLIST

Does it solve the task?

Does it handle the unexpected?

Can you trust the outcome?

LESS GUESSWORK. BETTER BUILDING.FIG. 01

THE 30-SECOND TAKE

01Learn to evaluate AI.

02Keep your foundations strong.

03Build something people can use.

A SIGNAL. A SKILL. A NEXT STEP.

Make sense of what’s next.

Find your field. Follow your curiosity. Leave with something to try.

Reviewed 19 Sep 2026

6 ideas to explore

AI & Data

This month’s focus

AI evaluation is a building skill.

An impressive answer isn’t the same as a reliable product. Learn to test the quality of an AI system before someone depends on it.

AI evaluationPythonTest design

Why it matters & what to build

Anthropic’s engineering guide describes evaluating both an agent’s actions and its final result. Our take: knowing how to define and measure success is a useful next step after building your first AI feature.

TRY IT YOURSELF

Build a tiny evaluation suite for an AI assistant.

Choose one task and write 10 realistic examples, including tricky cases.
Define what a good answer looks like before running your tests.
Compare two prompts and document where each one fails.

Explore the related course

Build on the basicsAnthropic

Engineering

Adoption signal

TypeScript earns its place in your toolkit.

As AI helps write more code, being able to read, check, and maintain that code becomes an especially useful habit.

TypeScriptAPI designTesting

Why it matters & what to build

GitHub’s 2025 Octoverse report places TypeScript first by monthly contributors in August 2025. That measures activity on GitHub, not job vacancies. Our take: pair type safety with runtime validation and tests.

TRY IT YOURSELF

Turn a small JavaScript app into a safer TypeScript app.

Type the data returned by one API endpoint.
Handle loading, empty, and error states explicitly.
Add a runtime check and a test for an unexpected response.

Explore the related course

Build on the basicsGitHub

Product & Design

Foundation to keep

Good design includes the edge cases.

Keyboard access, clear focus states, and forgiving forms make a product easier to use. Bring them into your design process early.

AccessibilityInteraction designUX writing

Why it matters & what to build

WCAG 2.2 adds criteria covering areas such as focus visibility, target size, and accessible authentication. Our take: use these as concrete design prompts when reviewing an everyday user journey.

TRY IT YOURSELF

Redesign a signup flow so more people can finish it.

Try completing the current form with only a keyboard.
Specify focus, validation, and error-recovery states.
Check labels, contrast, and tap targets; explain your changes.

Explore the related course

Beginner friendlyW3C

Cloud & Security

Worth exploring

AI features need security thinking, too.

A model with access to your tools introduces new failure modes. Start with permissions, untrusted inputs, and safe outputs.

AI securityThreat modellingPermissions

Why it matters & what to build

OWASP’s guidance covers risks including prompt injection and excessive agency. Our take: a useful security review follows what a model can access and do, as well as what it can say.

TRY IT YOURSELF

Threat-model a support assistant before giving it tools.

List its data sources, tool permissions, and possible actions.
Write examples of malicious instructions hidden in retrieved text.
Design a human approval step for consequential actions.

Explore the related course

Build on the basicsOWASP

AI & Data

Foundation to keep

Your data foundations still do the heavy lifting.

Before the model or the dashboard comes the data. SQL, clean datasets, and clear questions remain a practical place to begin.

SQLPythonData analysis

Why it matters & what to build

The 2025 Stack Overflow survey reports Python adoption grew, with SQL also widely used by respondents. Our take: build confidence in getting a trustworthy answer from a small dataset before reaching for more complex tooling.

TRY IT YOURSELF

Answer a real question with a small public dataset.

Choose a dataset and write down one question you want to answer.
Clean missing values and use SQL to calculate a useful metric.
Present one chart with its limitations and your conclusion.

Explore the related course

Beginner friendlyStack Overflow

Engineering

Worth exploring

Start with a workflow. Add an agent with purpose.

Not every AI feature needs autonomy. Clear steps and small, well-defined tools are a useful starting point for a dependable workflow.

AI workflowsTool designAPIs

Why it matters & what to build

Anthropic distinguishes predefined workflows from agents that choose their own steps, and recommends starting simply. Our take: understand the tradeoff before adding autonomous decisions to your app.

TRY IT YOURSELF

Build a three-step research workflow with visible sources.

Split the task into retrieval, summarisation, and review.
Define the input and output of each step.
Add a failure path and decide where a person should review.

Explore the related course

Build on the basicsAnthropic

Your saved topics stay in this browser. No account needed.

CONTEXT YOU CAN CHECK

Curated with care.
Grounded in sources.

Each edition connects industry research, developer surveys, and technical guidance to a practical learning step. The recommendations are our editorial interpretation; they aren’t a hiring ranking.

Sources & how to read this edition

Published 19 Sep 2026. Sources may predate this issue: this is a monthly learning guide, not a list of this month’s releases. “Foundation” highlights lasting skills; “Worth exploring” suggests a direction to investigate.

Demystifying evals for AI agentsAnthropic · 9 Jan 2026 Octoverse 2025GitHub · 28 Oct 2025 Web Content Accessibility Guidelines 2.2W3C · 12 Dec 2024 recommendation Top 10 for Large Language Model ApplicationsOWASP · 2025 edition 2025 Developer Survey: TechnologyStack Overflow · 2025 survey Building effective agentsAnthropic · 19 Dec 2024

TURN CURIOSITY INTO CAPABILITY

Your next step starts with a little practice.

Find your learning path

Make sense of what’s next.

AI evaluation is a building skill.

Build a tiny evaluation suite for an AI assistant.

TypeScript earns its place in your toolkit.

Turn a small JavaScript app into a safer TypeScript app.

Good design includes the edge cases.

Redesign a signup flow so more people can finish it.

AI features need security thinking, too.

Threat-model a support assistant before giving it tools.

Your data foundations still do the heavy lifting.

Answer a real question with a small public dataset.

Start with a workflow. Add an agent with purpose.

Build a three-step research workflow with visible sources.

Curated with care.Grounded in sources.

Your next step starts with a little practice.

Curated with care.
Grounded in sources.