Who predicts the future best?
We went through three years of predictions by top indie makers (and Elon Musk, for fun), found the original posts, and checked what actually happened. Each call links to its source and the evidence.
Best predictors
Ranked by accuracy. Minimum 3 resolved predictions.
- 1
92% - 2
44% - 3
27% - 4
20% - 5
0%
Worst predictors
Lowest accuracy first. Bold calls are allowed to miss.
- 10%
- 220%
Not enough data yet
Fewer than 3 resolved predictions so far.
100%
100%
75%
75%
75%
50%
50%
50%
50%
13%
—
—
—
—
—
—
—
—
Accuracy by topic
Across every tracked maker.
- AI49% of 33
- Business55% of 10
- Space30% of 10
- Cars39% of 9
- Other38% of 6
- Tech38% of 6
- Politics0% of 1
- Society50% of 1
Calls that came true
MailCheap would reach $1K MRR in under two weeks if growth stayed linear, six times faster than DataFast.
TrustMRR lists MailCheap at about $2.5K MRR in early November 2026, well past the $1K target.
A native code sandbox would let TypingMind support new standards like Anthropic's agent skills.
TypingMind now has a Skills system that installs SKILL.md-based agent skills from GitHub, plus a public skills directory.
Stack Overflow is finished as AI replaces it for developer Q&A.
Monthly questions fell to 2009 levels by mid-2025 and dropped about 78% year over year by Dec 2025. The decline was already obvious when he posted, so this was low-risk.
Deep Research would soon come to TypingMind.
TypingMind ships a Deep Research plugin that plans, searches and reads the web to produce reports.
Starship launches again next month
Flight 10 launched Aug 26, 2025.
And next year, Falcon is aiming for >150 flights!
165 Falcon 9 launches in 2025.
He would build a small team to explore new products such as an email app.
In November 2025 he announced he was building a new email client.
Set a goal of building a software business worth $1M, defined as DataFast reaching $20K MRR.
He reported crossing $20K MRR about 512 days after setting the goal (March 2026); TrustMRR shows DataFast around $29K MRR by late 2026.
Calls that missed
New Roadster unveil hopefully next month. It will be a banger next-level.
Unveil slipped to an Oct 2026 event.
Starship flies again next month
Slipped to May 22, 2026.
Grok 5, a roughly 6-trillion-parameter model, would arrive very soon, around the end of 2025 or early 2026.
Grok 5 missed both the Q1 and Q2 2026 windows; xAI shipped Grok 4.x updates instead and as of August 2026 Grok 5 still had no release date.
LLMs plateaued after GPT-4 and won't make big further leaps.
Reasoning models, GPT-5 and Gemini 3 (Nov 2025) all brought large benchmark and capability gains well beyond GPT-4.
Grok 5 will be out before the end of this year and it will be crushingly good
Grok 5 still not released as of Nov 2026.
AI agents become way more powerful once they can read/write to a database. I predict ChatGPT/Claude/etc will add Airtable-like databases within the next 2 months. It makes working with ANY type of data so much bette...
Neither ChatGPT nor Claude added an Airtable-like database product; they integrate with Airtable instead.
Tesla would probably offer autonomous ride-hailing to about half of the US population by the end of 2025.
At the end of 2025 Tesla ran small robotaxi fleets in a handful of Texas and California areas, a tiny fraction of the US population.
Pivoting Maybe to natural-language B2B forecasting tools would generate revenue from day one, starting by the end of August 2025, and keep the company alive.
In March 2026 he announced Maybe had been shut down at the start of the year; none of its directions, B2B finance included, found traction.
Still open: watch these
Not resolved yet, so not scored.
I think actors will become obsolete in 5 years.
Not resolved yet.
everyone likes to think they can detect AI ("it just has a certain look" or "it sounds like AI" or "humans don't write like that") but by this time next year absolutely nobody will be able to tell. nobody. not a singl...
About 1-year horizon.
New 2028 prediction: You work for the AI Cartel - or you're not in business.
Not resolved yet.
Looks like we will probably catch the ship with the tower in a few months. If there had been a tower out to sea where we practiced landing the ship, it would have been caught. First reflight of the ship will be eith...
Ship catch is targeted for Flight 15 (mid-Nov 2026).
I only build 2 types of startups now: 1) AI-proof in 10 years. My startup marketplace (TrustMRR) can be cloned with a prompt, but a network of 20K buyers can’t. Trust compounds over time. 2) Trends. Anything hot on ...
10-year horizon.
I don't think many humans are going to be reading or writing code in 5 years, so I don't really think it matters.
Not resolved yet.
His portfolio of small bets was on track to make $1M again in 2026.
June 2026 alone was about $84K; full-year total not yet reported.
Starlink coming to @AmericanAir next year!
Not resolved yet.
Former makers
75%
50%
—
—
—
Frequently asked questions
How is prediction accuracy calculated?
Each resolved prediction scores 1 if it came true, 0.75 if mostly true, 0.5 if partly true and 0 if wrong. A maker's accuracy is the average, shown as a percent. Pending predictions are listed but not scored, and a maker needs at least 3 resolved predictions to be ranked.
Where do the predictions come from?
From makers' public posts on X over roughly the last three years, plus podcasts and interviews when a post isn't available. Every prediction links to its original source and every verdict links to the evidence behind it.
Why is Elon Musk on an indie maker site?
Just for fun. He acts more independently than most solo founders, and his public timelines make him the most-tracked predictor on the internet, which gives indie makers a useful benchmark.
Is a prediction wrong if it came true late?
If a prediction named a date and missed it, it counts as wrong or partly true depending on how close it got. Predictions without a date are judged on whether they have happened by now; ones that still could are marked pending.