Daniel Hjermitslev

Predictions · scored by Jev

What I think happens next

Predictions about AI and frontend, each with a deadline and a way to check it. Jev, a judgment model, gives its read on how plausible and how bold each one is. When a deadline passes, I mark it right or wrong.

Each card has two readings from Jev. Plausible runs from “against the trend” to “near certain” by the deadline; bold runs from “consensus” to “radical”. The brighter the blocks, the more Jev leans that way.

2027

7 predictions
  1. By 30 June 2027

    Shopify, WooCommerce and DanDomain all offer an official MCP server, so AI agents can work directly with a shop's data.

    Counts as true if

    Each of the three platforms has published an official, documented MCP server for merchants.

    Jev's confidence: plausible 0.63 · bold 0.72

  2. By 31 December 2027

    "Design engineer" becomes a common job title for software roles in Denmark.

    Counts as true if

    Jobindex lists at least 50 software (not mechanical or civil) design engineer postings in 2027.

    Jev's confidence: plausible 0.43 · bold 0.66

  3. By 31 December 2027

    Chrome, Safari, Firefox and Edge all ship a built-in language model that web pages can call.

    Counts as true if

    All four browsers expose an on-device language model API to web pages in their stable release.

    Jev's confidence: plausible 0.46 · bold 0.52

  4. By 31 December 2027

    AI agents complete at least 10 percent of orders on Danish webshops on behalf of shoppers.

    Counts as true if

    Dansk Erhverv, FDIH or a major payment provider reports that at least 10 percent of Danish online orders in 2027 were placed by an AI agent.

    Jev's confidence: plausible 0.69 · bold 0.69

  5. By 31 December 2027

    At most of Denmark's ten largest web agencies, AI agents write most new frontend code, and people review it.

    Counts as true if

    At least six of the ten largest Danish web agencies, ranked by employees in the Danish business register (CVR), state publicly that AI agents write most of their new frontend code.

    Jev's confidence: plausible 0.65 · bold 0.69

  6. By 31 December 2027

    AI literacy courses become a standard offer at Danish jobcentres.

    Counts as true if

    The Danish Agency for Labour Market and Recruitment (STAR), Local Government Denmark (KL) or a national survey reports that at least half of Denmark's 98 municipalities offer AI literacy courses to unemployed citizens through their jobcentre.

    Jev's confidence: plausible 0.54 · bold 0.22

  7. By 31 December 2027

    "Prompt engineer" has almost disappeared as a job title, absorbed into ordinary roles.

    Counts as true if

    Jobindex lists fewer than half as many Danish postings titled prompt engineer in 2027 as in 2024.

    Jev's confidence: plausible 0.41 · bold 0.59

2028

3 predictions
  1. By 31 December 2028

    A typical Danish small business has at least one AI agent running unattended on a routine task, such as bookkeeping or the inbox.

    Counts as true if

    Statistics Denmark or the Danish Business Authority reports that more than half of small businesses use an AI agent that acts without a person approving each step.

    Jev's confidence: plausible 0.65 · bold 0.59

  2. By 31 December 2028

    Most AI features in Danish public-sector software run on models hosted by European providers or on the device, not by US cloud providers.

    Counts as true if

    The Danish Agency for Digital Government, Local Government Denmark (KL) or the National Audit Office (Rigsrevisionen) reports that most AI features in public-sector systems run on models hosted by European providers or on the device. US providers' EU regions count as US.

    Jev's confidence: plausible 0.40 · bold 0.27

  3. By 31 December 2028

    Job postings for junior frontend developers in Denmark are at least 30 percent below their 2024 level.

    Counts as true if

    Jobindex data shows at least 30 percent fewer junior frontend postings in 2028 than in 2024.

    Jev's confidence: plausible 0.46 · bold 0.18

2030

2 predictions
  1. By 31 December 2030

    Glasses cut the cord. By the end of 2030, Apple, Google or Meta sells AR glasses that work without a phone at all, with their own connection and their own apps.

    Counts as true if

    Apple, Google or Meta sells consumer AR glasses with their own cellular or Wi-Fi connection that can be set up and used, including third-party apps, without pairing to a phone or computer. Meta's Ray-Ban Display (2026) runs third-party web apps but needs the phone app for setup and connection, so it does not count.

    Jev's confidence: plausible 0.66 · bold 0.51

  2. By 31 December 2030

    Glasses get their smartphone moment. Phones outsold PCs four years after the first iPhone; by the end of 2030, the world buys more AR glasses than tablets.

    Counts as true if

    IDC, Counterpoint or a comparable analyst reports that more AR glasses than tablets were shipped worldwide in 2030. Headsets such as Apple Vision Pro or Meta Quest do not count.

    Jev's confidence: plausible 0.62 · bold 0.70

2034

1 prediction
  1. By 24 January 2034

    Since the Macintosh in 1984, computing has meant windows, a mouse and a screen. By its 50th birthday, the screen starts moving into the eye: a million people buy AR contact lenses in a single year.

    Counts as true if

    IDC, Counterpoint, a comparable analyst or the maker reports that AR contact lenses with a built-in display were sold to more than one million consumers worldwide in 2033.

    Jev's confidence: plausible 0.83 · bold 0.71

How the scoring works

Jev is a judgment model from TypeSafe. It does not write or reason in steps: it reads a situation and returns a probability for each answer I define. That makes it a snap read, not a forecast.

Each prediction gets the same three questions. The pixel lines show Jev's whole distribution, so a prediction it is torn on looks torn.

The questions Jev answered

  1. Judging from what was publicly known as of `as_of`, how likely is it that the claim in `prediction.claim` is true by `prediction.deadline`, as decided by `prediction.resolves_true_if`?

    It contradicts clear current trends and would need a reversal to happen / It is possible, but several unlikely things would have to happen first / Current trends point this way, but it could easily miss the deadline / The trend is strong and the deadline leaves enough time / It is already largely true, or close to certain by the deadline

  2. As of `as_of`, how far is the claim in `prediction.claim` from what most frontend and AI practitioners already expect?

    Most practitioners already take it for granted / Many practitioners expect it and few would argue / Practitioners are split on it / Most practitioners would doubt it / Most practitioners would dismiss it outright

  3. Could a neutral person decide on `prediction.deadline` whether the claim in `prediction.claim` came true, using only the condition in `prediction.resolves_true_if`?

Scored with jev-1.13.0 on 2026-09-24. Nothing runs when you load this page.