Grok in the Timeline · Case Study
Live · June 6, 2026
For the Exceptional Designer role · xAI · Palo Alto

Grok shouldn't be a tab. It should be a layer.

𝕏 9:41 ●●● 100%
For you Following
Karpathy @karpathy · 2h
Scaling reasoning at inference: depth vs width is a clean separation nobody is exploiting yet.
Analyze with Grok
Summarize this
Is this true?
Explain like I'm five
Open in Grok
The thesis, in one tap.
Top-right of every post Tooltip menu on tap Anchored, never a destination

The Grok glyph lives in the post header, exactly where X shipped it. One tap opens a tooltip with the four shortcuts every post is hired for: Summarize, Is this true?, Explain, Open in Grok.

A pixel-level argument for treating Grok as ambient infrastructure inside the timeline - not a destination users have to walk to. Six surgical insertions, ordered by impact. One monochrome system. Zero new tabs.

What this gives you: a designer who already ran the thesis, the visual board, the working prototype, the 90-day plan, and the rollout gates - so the first conversation can be about what to ship, not whether the candidate gets it.

Let's talk 5 min? Open the prototype Or read the pitch
Format
Case study + working prototype
Design
x.com timeline, grok.com chrome
If hired · 90 days

What I'd ship.

Weeks 1-2
Listen and instrument. Sit with eng on the model team and the X timeline team. Pull current Grok telemetry - dismiss rate, hallucination report rate, latency p50/p99. Write down the actual numbers.
Weeks 3-6
Ship Insertion 1: Inline Ask-Grok. Smallest unit that proves the layer-not-tab thesis. Forces resolution on latency budget, anchoring pattern, source-chip language. Behind a 0.5% flag on Premium+.
Weeks 7-10
Ship Insertions 3 + 4 in parallel. Claim-Context and Compose Co-Writer. Different surfaces, no shared dependencies. Co-Writer is the conversion lever for daily post-creation.
Weeks 11-13
Write the design-system extension for Grok-in-product surfaces. One mark, one inner-border treatment, one source chip. Defended in writing so the next five designers don't reach for purple.
Counter-week
What I'd refuse to ship in 90 days: a Grok sidebar in the timeline, a verdict-rendering fact-check feature, a settings page that's anything but a single-tap dismiss. The IC's job is to make the case against them with evidence.
Live · Interactive

Try the prototype

The working product, a near pixel-exact x.com/home. Tap the corner Grok button on any post and watch the answer stream in. Live ticker, breaking-posts pill, thread synthesizer.

Launch prototype
Flagship · Connects real X

Grok Organize

The bookmark organizer. Connect a real X account over OAuth, and Grok clusters your saves into themed collections so a graveyard of bookmarks becomes a two-minute review.

Open Grok Organize
Start here · 3 min

The breakdown

One page that maps the whole proposal. What it is, who it's for, what each page argues, with a live demo in every themed section.

Read the overview
Latest · V1.5 · June 5, 2026

The flagship shipped: Grok Organize.

Grok Organize connects a real X account over OAuth, reads your bookmarks, and clusters them into themed collections so a saved pile becomes a two-minute review. The prototype is now a near pixel-exact x.com/home, X's real recommendation algorithm is documented and wired into how saves get ranked, and all 26 pages now share one navigation spine.

Open Grok Organize → How the algorithm ranks Read the master file V1.0 to V1.5 timeline
04 / System

Six principles, no exceptions

  1. 01

    Grok is a layer, not a destination

    Every Grok affordance lives inside the timeline, post detail, or compose. No new top-level tab. The destination model is the failure mode this whole spec is structured to avoid.

  2. 02

    One tap to invoke, zero to dismiss

    Grok answers expand inline below the post that prompted them. Scrolling away dismisses. No modals. No close buttons. No state to manage.

  3. 03

    Always cite. Always show confidence.

    Every Grok-rendered claim carries source chips and an explicit confidence indicator. Hidden uncertainty is dishonest. Showing it builds the calibration loop users need.

  4. 04

    Context, never verdict

    Grok surfaces underlying data. Users draw conclusions. Community Notes already does fact-checking with strong governance - that's the right tool for that job.

  5. 05

    Monochrome is the brand position

    One Grok mark, one inner-border treatment, one motion curve. No purple. No gradients. No "AI shimmer." The restraint is the point - X earns its severity, and Grok earns its place by respecting it.

  6. 06

    Latency is a feature, not a metric

    400ms first token. 1.5s full answer. If we can't hit budget on a surface, we don't ship that surface. Slower than scrolling is dead on arrival.

05 / Scope

Six insertions, ranked

01

Inline Ask-Grok

One-tap "what's this about" answer anchored to any post. The cell expands; the user never loses their place.

Impact: very high
Effort: medium
02

Thread Synthesizer

Consensus / disagreement / notable summary above any thread with ≥8 replies. Visibly generated, collapsible, sample-size disclosed.

Impact: very high
Effort: high
03

Claim-Context Chip

Quiet "context available" affordance on posts with checkable claims. Context, never verdict. Sources first.

Impact: high
Effort: medium
04

Compose Co-Writer

Private draft review - sharpen, steelman, check. The lurker participant conversion lever.

Impact: very high
Effort: medium
05

Steerable For You

Adjust your algorithm in plain language. Diff before apply. The moat.

Impact: very high
Effort: high
06

Conversational Time-Scrub

Query X's history as a real timeline, filtered to your interest graph. Ship after 1–4 prove out.

Impact: medium
Effort: very high