Frontend developer

A US federal rule binding state and local governments under Title II of the Americans with Disabilities Act. The 2024 rule requires their web content and mobile apps to meet WCAG 2.1 AA; this interim final rule, effective 20 April 2026, extends compliance to 26 April 2027 for entities with populations of 50,000 or more and to 26 April 2028 for smaller ones and special districts. In its reasoning the Department says advanced technology such as generative AI does not yet reliably automate the remediation of inaccessible content at scale, and that the less public entities can rely on technology to make content accessible, the more they will need to rely on manual work instead. That is the Department's reasoning, not an operative requirement, and it is about fixing existing content rather than building new interfaces; the rule binds governments, not private companies.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

The filing establishes metering, not its effect. Figma states it introduced AI credits across all seats during 2025 and began enforcing the limits in March 2026, selling monthly add-ons or pay-as-you-go usage beyond them; the filing names the month, not a day. It reports no usage figure for Figma Make or Dev Mode. It says only that, after enforcement began, it saw more support requests, some public dissatisfaction and reduced usage by certain customers, with no measurement of how much. Figma sells these seats, so it has an interest in how a pricing change is presented — but the fact recorded here, that generating an interface from a design now costs metered credits rather than being included in the seat, is stated in a filing where overstating carries liability. It says nothing about tools from other vendors, and nothing about in-house pipelines.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

The UK central government's standard for building public digital services. If a service uses AI, it must make clear to users how this might affect their data and the outcome or information they receive; if it uses an AI-powered chatbot, it must make sure users understand that the answers do not come from a human and might not be accurate, and know how to contact a human being. The page also says services should be prepared to stop using AI if needed. It binds UK central-government service teams, not private companies, and it sets what the interface must communicate, not how any team builds it.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

The US consumer-protection regulator's complaint against a company that sold an AI-powered widget said to make any website compliant with the Web Content Accessibility Guidelines, settled by a final order requiring a $1 million payment and barring such claims without evidence. The complaint states that, according to accessibility experts, no automated testing tool alone can determine whether a website meets accessibility standards and manual human testing is required, and lists barriers that remained on sites using the product: missing focus indicators, keyboard traps in forms and menus, missing or inaccurate alt text, incorrect headings. It is a deceptive-advertising case settled by consent, the expert sentence cites outside experts, and the affected sites are not named.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

One company's own codebase, published by that company, with the hand-migration estimate (1.5 years of engineering time) supplied by the same company and not checkable from outside. What makes the result reproducible is the mechanism, which the post describes in full: the migration is a state machine whose every step has an automatic referee - jest, eslint and tsc say pass or fail - so the model could retry without a person watching, and the first bulk run cleared 75% of files in four hours. The remaining tail is the more useful number: after four days of tuning, 97%; many of the long-tail files had been retried between 50 and 100 times, and the last 3% were finished by hand. The post is also explicit that the main driver was selecting the right related files to put in the prompt - which grew to 40,000-100,000 tokens pulling in up to 50 files - rather than prompt wording. It reports no change to headcount, hiring or roles, and does not claim one.

An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.

The UK central government's standard for building public digital services. All government services must follow progressive enhancement — work with HTML first, then add CSS and JavaScript — even if part of the service needs JavaScript. To make JavaScript work across browsers, the page lists feature detection, polyfills and transpiling, using automated tests or linters, and doing regular manual testing, including testing with older or lower-powered devices. It binds UK central-government service teams; it is a standard for how the work must be done, not a measurement of how much of it is automated.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

The task you would previously have been hired to do first — turning a design into a screen — is the one most exposed here. That is worth knowing before you spend a year getting fast at it. The parts that hold are the states nobody drew and the requirement that it work for everyone, and neither is taught by building portfolio pages.

Your leverage moved from producing screens to constraining them: the design system, the states, the performance budget, the accessibility obligation. That is less visible work and harder to show in a portfolio, which is a real problem when you next change jobs — start keeping a record of what you refused and why.

Stay and strengthen

Own the constraint layer

When screens become cheap, the scarce thing is the person who keeps a product coherent — the component library, the states, the performance budget.

Reshape the role

Become the one who can prove it is accessible

This is moving from good practice to obligation in several markets, and an obligation needs a named person who can demonstrate compliance rather than assert it.

Adjacent move

Move toward the product decision

Frontend engineers see what users actually do more directly than almost anyone, and that observation is the raw material of product work.

No number here would be honest, and the question is better asked task by task. Producing a screen from a design is already largely done by machine; deciding what the screen does when the data is missing is not, and neither is being answerable for whether it works for someone using a screen reader. A signal you can check yourself: how much of your last month was spent typing markup versus deciding behaviour. If it is mostly the former, the exposure is real and the move is toward the latter.

On its most visible task, yes, and the reason is mechanical rather than a judgement about difficulty: the input is visual, the output is text, correctness is checkable by looking, and the training data is the public web. But most of a real frontend codebase is not that task — it is the states nobody drew, the devices nobody tested on, and the constraints that keep a growing product coherent. Exposure on the visible half is not the same as exposure on the job.

Breadth is not automatically protection, and that is the framing worth dropping. Two shallow exposures are easier to replace than one deep one; what protects is being the person who can be trusted to judge, and judgement is domain-specific. If you cross over, cross for a reason — the states and the data model are genuinely connected, and understanding both makes you better at the boundary, which is where most bugs live.

It is still asked for, and it is a weakening signal for the same reason the first task on this page is: a reviewer now knows what a good-looking screen costs to produce. What has not weakened is evidence of judgement — a component you removed and why, a state you handled that nobody specified, a performance budget you held against pressure. Those are harder to assemble and much harder to generate.