47 Tech Predictions, 18 Months Later: The Public Scoreboard

I made 47 tech predictions in February 2024. Eighteen months later, here is the scoreboard with receipts. I got 28 right (60%), 9 wrong (19%), and 10 are still pending. The pattern of what I got right and what I got wrong is more interesting than the numbers.

Most tech predictions are click-bait with no accountability. The pundits who got 2023 wrong are publishing 2027 predictions today, with no acknowledgement of the misses. I find that intellectually dishonest. So I logged my own predictions publicly, dated them, and committed to scoring them out loud. This is the first eighteen-month checkpoint.

The scoring methodology

Three categories, scored against a fixed rubric agreed before the data came in:

• Right. The predicted outcome happened, in roughly the predicted shape, within the predicted timeline. Half-credit if the direction was right but timing slipped by less than 12 months.
• Wrong. The predicted outcome did not happen, or the opposite happened, or the timeline has slipped so far that the prediction is functionally dead.
• Pending. Not enough evidence to score yet. Default position when the underlying market signal is mixed.

Predictions are scored against the public version, not the steel-manned version I would write today. That is the point of accountability.

The hits (6 strongest)

1. Passwordless becomes the enterprise default by late 2025

Predicted: passwordless authentication (passkeys, FIDO2, WebAuthn) crosses the line from "interesting" to "default for new enterprise rollouts" by Q4 2025. Outcome: confirmed. Apple, Google, and Microsoft all defaulted to passkey enrolment by mid-2025, and the enterprise vendors followed. The supporting analysis lives in passwordless everything.

What made it predictable: the device ecosystem had already shipped the primitives. The blocking constraint was UX, not protocol. Once Apple committed, the rest fell.

2. Phishing-resistant auth becomes a compliance default

Predicted: OMB and CISA pressure pushes phishing-resistant MFA from "best practice" to compliance default within 18 months. Outcome: confirmed. The federal mandate accelerated, and the private sector compliance frameworks (PCI DSS 4.0, NIST 800-63-4) followed. Full detail in phishing-resistant auth default.

What made it predictable: regulatory direction is the most stable signal in cybersecurity. When the federal procurement engine commits, the private sector follows within two budget cycles.

3. Machine identity volume overtakes human identity

Predicted: the count of non-human identities (service accounts, workloads, agents) in the average enterprise exceeds human identities by mid-2025. Outcome: confirmed and then some. The ratios at large enterprises now run 50:1 or higher. Background in the machine identity explosion.

What made it predictable: every move toward microservices, multi-cloud, and agentic systems generates non-human identities by the dozen. The base rate of growth was already exponential in 2023.

4. Coding in English becomes a mainstream developer skill

Predicted: natural-language-first coding moves from demo to default developer workflow within 18 months. Outcome: confirmed. Cursor, Claude Code, and Codex have made prompt-first coding the default for a measurable share of professional developers. Full prediction.

What made it predictable: the productivity delta was visible by late 2023. Once the delta crosses 2x, adoption is a forcing function, not a choice.

5. The autonomous SOC becomes a real product category

Predicted: "autonomous SOC" graduates from marketing slogan to a venture-funded product category within 18 months. Outcome: confirmed. Prophet, Dropzone, Torq, and the incumbents all shipped recognisable autonomous SOC products. Detail in autonomous SOC.

6. Zero trust becomes the default network posture for new deployments

Predicted: zero trust crosses the line from initiative to default for new enterprise deployments by mid-2025. Outcome: confirmed. New deployments default to identity-aware proxies, microsegmentation, and explicit verification. Background in zero trust default.

The misses (6 wrong ones)

The misses are where the analysis gets useful. Each one came with a bias I can now name.

1. Self-driving cars as default new-vehicle option by 2026

Predicted: SAE Level 4 becomes a default option on new mass-market vehicles by 2026. Outcome: wrong. Waymo and Zoox are operating, but the consumer vehicle default is still SAE Level 2+. The regulatory and insurance bottlenecks were harder than the engineering. Detail in self-driving cars.

The bias: timeline compression on technology where the unsolved layer is institutional, not technical.

2. Robotaxis as default urban transport by 2026

Predicted: robotaxis displace ridesharing as the default in 10+ US cities by 2026. Outcome: wrong. Waymo has expanded, but the displacement is modest. Full prediction.

The bias: same as above. I underweighted the cost-per-mile gap and the regulatory drag.

3. AR glasses as the post-smartphone default by 2026

Predicted: AR glasses cross the line from prosumer toy to mainstream daily-driver hardware by 2026. Outcome: wrong. The hardware is closer, the form factor is not. Meta's Orion and Apple's Vision Pro shipped, but neither is a smartphone replacement yet. Full prediction.

The bias: dominant-narrative anchoring on Meta and Apple roadmaps that turned out to be more aspirational than reported.

4. Personal AI clones as a mass-market product by 2026

Predicted: voice and avatar clones move from creator-tool curiosity to mass-market identity product by 2026. Outcome: wrong. The technology shipped. Consumer adoption stalled on the trust and identity-fraud questions, which I underweighted. Full prediction.

The bias: I missed the second-order regulatory and reputational backlash.

5. The cloud IAM monopoly cracks by 2026

Predicted: AWS, Azure, and GCP IAM models face credible third-party competition by 2026. Outcome: wrong. The competition exists. The default deployment did not move. Full prediction.

The bias: I overweighted the technical case and underweighted the lock-in moat.

6. Agentic browsing becomes the default research pattern by 2026

Predicted: AI agents that browse and synthesize on your behalf become the default research pattern for knowledge workers by 2026. Outcome: half-wrong. The technology is real and adoption is growing, but the default has not shifted. Full prediction.

The bias: timeline compression again. The capability arrived. The behaviour change is slower.

The four patterns of prediction failure

Looking at my own misses and at the misses of other people I respect, four patterns recur. I am now using them as a checklist before I publish a new prediction.

1. Timeline compression

The most common failure mode. The technology will ship. The adoption curve will take longer than the technologist expects, because adoption is a function of incentives, switching costs, and habit, not of capability. Three of my six misses fall into this bucket.

2. Dominant-narrative anchoring

When Meta says Orion is shipping, the prediction drifts toward Meta's roadmap. When OpenAI says agents are six months away, the prediction drifts toward OpenAI's roadmap. The fix is to discount vendor-stated timelines by 50% as a default.

3. Missing a regulatory variable

Predictions that ignore the regulatory layer are predictions about a fantasy world. Self-driving cars, AI clones, and crypto adoption have all moved more slowly than the technology because the regulatory layer was the binding constraint.

4. Second-order effect blindness

The first-order prediction (LLMs will replace search) is the easy call. The second-order question (what does that mean for content marketing, SEO budgets, B2B funnels) is the harder call. I have started writing the predictions and the second-order implications in the same document. The implication thread is often where the actual bet lives. I tried this discipline in my analysis of AI-powered search and in my broader piece on the future of search.

The cross-portal context

For predictions that resolved as obituaries (the tech that died) rather than predictions that came true, the tech graveyard portal tracks the deaths. The asymmetry is instructive: predicting what dies is easier than predicting what wins, because the death of a technology has fewer moving parts than the adoption of one.

The predictions I am most uncertain about for 2026-2027

Closing on epistemic humility. The five 2026-2027 predictions I would not bet on with confidence:

1. Post-quantum crypto rollout speed. NIST has published the standards. Adoption is gated on hardware and tooling. The compliance forcing function will arrive. I do not know when. Context in post-quantum crypto rollout.
2. Personal AI agents transacting on the user's behalf. The capability is shipping. The trust model is not. Full prediction.
3. The future shape of CIAM. I have argued legacy identity systems are dead. The shape of the replacement is still uncertain.
4. Edge AI compute fabric maturity. Detail here.
5. Behavioural biometrics as a default authentication layer. Detail here.

The meta-commitment

I am going to keep doing this. Annually. With receipts. The point is not to be right, it is to be calibrated. A 60% hit rate is fine for an 18-month checkpoint, as long as the misses are honest and the patterns are named. The predictions and the scoreboards live at the future tech portal.

If you publish predictions, score them publicly. The pundit class has gotten away with no accountability for too long.