01 · 15 years at a glance
From Symfony to Claude Code: 15 years on X
02
Activity
My peak was 2016: 1,359 tweets. Since 2022, I've settled at about 300 tweets and 600 likes a year.
Items per month by type of action, Feb 2011 → Sep 2026. Likes are dated by the liked tweet's creation date (a lower bound).
Scroll the chart sideways →
Show the data (per year)
Retweeting became my main move: 74–75% of my tweets in 2025–2026, vs 45% in 2015.
Share of retweets among my tweets (original + replies + retweets), per year. 2026 ends on Sep 29.
Show the data
03
Topics over time
Click a topic to follow it across every chart on the page; click it again to reset.
In three years, AI went from 1% to half of what I read and write: 51% in 2025, vs 1% in 2022.
Share of each topic per quarter, all actions combined (likes, tweets, replies, retweets). The three AI topics sit at the bottom, in purple.
Scroll the chart sideways →
Show the data (per year)
PHP/Symfony, my #1 work topic every year until 2020, drops to 11th in 2026; coding with AI agents takes the top spot.
Rank of each topic per year (1 = most frequent, personal included). Four topics are highlighted; hover a line or pick a topic to follow it.
Scroll the chart sideways →
Show the data
My range of topics kept widening until 2024 (3.31 bits), then narrowed in 2025 (3.12) as AI took over.
Topic diversity per year: Shannon entropy of the split across the 14 topics (about 3.8 bits would mean all 14 equally present).
Show the data
04
Life cycles
Only my AI topics are at their peak right now; PHP/Symfony peaked in late 2017 and is still declining.
One row per topic, sorted by peak date. Area: items per quarter (each row has its own scale). Thick bar: sustained period (at least 3 items, in 2 out of 3 consecutive quarters). Dot: peak. Right: last 12 months (Oct 2025–Sep 2026) vs the previous 12.
Scroll the chart sideways →
Show the data
05
Reading vs writing
I write less and less: writing went from 60% of my actions in 2015 to 12% in 2026.
Writing share = tweets I wrote (original + replies) ÷ (tweets I wrote + likes), per year. Retweets excluded.
Show the data
Infra and Cloudflare is the tech topic I write about most today: a lift of 2.5 to 2.7 every year since 2024.
Lift = a topic's writing share ÷ my overall writing share that year. Red: I write about it more than my average; blue: I mostly read. Rows sorted by recent lift. Empty cells: fewer than 10 items. Measured from 2014 (too few likes before).
Scroll the chart sideways →
Show the data
On AI, writing follows reading: a 4-quarter lag for agentic coding, 11 quarters for AI models.
First quarter of sustained reading (likes) and of sustained writing (tweets + replies), per topic. Before 2014, likes are too sparse to date reading, so negative lags there are an artifact.
Scroll the chart sideways →
Show the data
AI fills my likes, not my tweets: 685 likes vs 51 tweets of my own on the two main AI topics in 2024–2026.
Per topic and year: above the line, what I liked; below, what I wrote (original + replies). Same scale above and below, specific to each topic.
Show the data
06
What resonated
My product/design and career tweets land best: 51% and 40% get at least one like, vs 23% for coding with AI agents.
Share of my tweets (original + replies) with at least one like, per topic. Counts as of the export date. Faded bars: fewer than 30 tweets.
Show the data
Apart from one viral tweet in 2014, engagement stays modest: the median is 0 likes in almost every topic.
Likes received per topic and year on my own tweets (log scale). Hover a cell for tweets, likes and retweets.
Scroll the chart sideways →
Show the data
My 10 most-liked tweets: half are personal, and the top one (Nov 2014, 2,891 likes) dwarfs the rest.
Ranked by likes received. Tweet texts and links are deliberately left out.
07
Languages
English took over my likes (37% in 2014, 81% in 2025); my own tweets followed in 2025 (82%) before French came back in 2026.
Share of English per year, by action. Likes: heuristic (some remain undetermined); my tweets: language detected by X. Likes shown from 2013 (fewer than 10 before).
Show the data
08
Sources
github.com ruled 2014–2017 (244 links); since 2018, my top “domain” is x.com, i.e. links to other tweets.
Most-shared domains per era (original tweets, replies, retweets). Same scale for every era. Liked tweets' domains aren't in the archive.
Link shorteners are gone: 42.5% of my links in 2011, 0% since 2023.
Share of links shared through bit.ly, goo.gl, buff.ly and other shorteners, per year.
Show the data
Each era has its domains: elao.com early on, commitstrip in 2015, welcometothejungle since 2021.
Links shared per domain and year, sorted by average year of use.
Scroll the chart sideways →
Show the data
09
Vocabulary
My tech eras show up in the words that emerge: Ansible, Vue.js, serverless, Cloudflare, ChatGPT, then Claude Code and MCP.
Per year, terms whose rate per 1,000 items is at least 3× that of the previous 3 years (8 occurrences minimum). Right: the multiplier. Bold: era terms (25+ occurrences, mostly tech). Terms appear as written, in French or English; a few personal terms were removed.
Every technology has its wave: Symfony2 peaks in 2014, serverless in 2019, Core Web Vitals in 2023, Claude in 2026.
Frequency of tech terms per year (per 1,000 items), relative to each term's own maximum. Sorted by peak year. Right: total occurrences.
Scroll the chart sideways →
Show the data
I adopt at the crowd's pace: a one-year median between a technology's release and my sustained use of it; only INP came early.
For 14 terms: public release date (general-knowledge reference, approximate) and the first year of sustained use in my archive.
Scroll the chart sideways →
Show the data
10
Rhythm
My hours shifted: weekends emptied out and late nights grew.
My tweets (all types) by weekday and hour (local time), as a share of each era's total. Same color scale for all five eras.
Weekends went from 20% of my tweets in 2016 to 4.5% in 2026; late night (midnight–5 am), from 3% in 2011 to 11–14% in 2021–2025.
Share of my tweets posted on weekends and at night, per year.
Show the data
12
All 23 findings
Every finding in one sentence, grouped by part. Each group title links to its charts.
13
Caveats
- Like dates are lower bounds. The archive gives the liked tweet's creation date, not the date of my like, and like.js is only partly ordered (29% inversions). Before 2014 there are fewer than 80 likes a year, so lags and shifts are only measured from 2014.
- No bookmarks in the archive, and no domains for likes (like.js only contains t.co links).
- Topics come from an AI classifier. Every item was classified by Jev (jev-1.13.0, topic grid v2, 18,666 items): median confidence 0.93; 57% of items at 0.9 or above, 9.6% below 0.5; for 24% of items, the second-best topic has a probability of at least 0.2. 701 items with no usable text are left out of topics.
- Small recent volumes. I write 70 to 80 tweets a year now, so the 2024–2026 lifts rest on 10 to 40 tweets per topic.
- Engagement counts are cumulative as of the export date; the 2014 viral tweet crushes averages, hence medians and the share of liked tweets.
- 2026 is partial (until September 29). Public release dates are approximate, and “early” is not proof of expertise.
- Privacy. For this public version, account names, tweet texts and links were removed, and some personal details were merged into the Personal topic or removed.