eric zakariasson(@ericzakariasson)

and a really neat mobile feature: set it as your action button and grok bot is one press away!

eric zakariasson (@ericzakariasson) — Mon, 19 Oct 2026 21:46:52 GMT

and a really neat mobile feature: set it as your action button and grok bot is one press away!

Grok Bot@bot

Grok Bot is now one step closer.

Set it as your iPhone's Action Button and press it to talk with Bot.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬2🔄0❤️14👀1934📊3

⚡ Powered by xgo.ing

grok bot 0.71.2 is out - smarter notifications - files up to 250 mb (1 gb for video), 20 per messag...

eric zakariasson (@ericzakariasson) — Mon, 19 Oct 2026 21:46:49 GMT

grok bot 0.71.2 is out

- smarter notifications
- files up to 250 mb (1 gb for video), 20 per message
- forms arrive pre-filled

and a lot more!

x.ai/changelog/bot

💬67🔄42❤️764👀25948📊179

⚡ Powered by xgo.ing

which connectors do you want to see next in grok @bot? some of the most requested right now are - a...

eric zakariasson (@ericzakariasson) — Thu, 15 Oct 2026 22:20:03 GMT

which connectors do you want to see next in grok @bot? some of the most requested right now are

- apple (icloud mail, calendar, reminders, imessage)
- whatsapp
- bitwarden
- apple health
- meta ads

what else?

eric zakariasson@ericzakariasson

grok @bot marketplace is open for submissions!

send us your connectors:
x.ai/bot/marketplac…

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬1119🔄41❤️1371👀92206📊1070

⚡ Powered by xgo.ing

grok @bot marketplace is open for submissions! send us your connectors: https://t.co/bkn3313Wjl

eric zakariasson (@ericzakariasson) — Wed, 14 Oct 2026 22:08:20 GMT

grok @bot marketplace is open for submissions!

send us your connectors:
x.ai/bot/marketplac…

Your browser does not support the video tag.

🔗 View on Twitter

💬151🔄69❤️1330👀271201📊495

⚡ Powered by xgo.ing

try it out npm i @xai-official/sdk https://t.co/KI6rjgbCx6

eric zakariasson (@ericzakariasson) — Wed, 14 Oct 2026 15:44:05 GMT

try it out

npm i @xai-official/sdk

github.com/xai-org/xai-sd…

💬0🔄0❤️9👀1825📊3

⚡ Powered by xgo.ing

and the full cookbook here: https://t.co/gaT92x8voP

eric zakariasson (@ericzakariasson) — Wed, 14 Oct 2026 15:44:03 GMT

and the full cookbook here: docs.x.ai/cookbook

💬3🔄1❤️24👀3767📊15

⚡ Powered by xgo.ing

cookbook is updated with 25 examples, all built on the @spacexsi typescript sdk: - short films from...

eric zakariasson (@ericzakariasson) — Wed, 14 Oct 2026 15:44:00 GMT

cookbook is updated with 25 examples, all built on the @spacexsi typescript sdk:

- short films from a one-line premise
- one photo to a time-lapse
- talking photos
- product photo to video ad
- video dubbing
- live voice translation

and more!

Your browser does not support the video tag.

🔗 View on Twitter

💬15🔄11❤️207👀13963📊77

⚡ Powered by xgo.ing

@bot @mail it's one email address per account (for now), shared by all your bots. sorry for the conf...

eric zakariasson (@ericzakariasson) — Tue, 13 Oct 2026 18:47:09 GMT

@bot@mail it's one email address per account (for now), shared by all your bots. sorry for the confusion!

💬16🔄5❤️126👀13460📊33

⚡ Powered by xgo.ing

grok @bot 0.70.2 is rolling out! - each bot can claim its own email address at @mail.grokbot.com an...

eric zakariasson (@ericzakariasson) — Tue, 13 Oct 2026 17:17:53 GMT

grok @bot 0.70.2 is rolling out!

- each bot can claim its own email address at @mail.grokbot.com and draft emails for you to send
- apps like Figma, Notion and Canva show up as live cards right in the chat
- bots can search X and read public posts and profiles without connecting an X account
- describe what you want and your bot suggests a matching template
- shared pages can be downloaded as a pdf
- an approval check before anything irreversible

and much more!

x.ai/changelog/bot

💬199🔄113❤️1875👀5526156📊550

⚡ Powered by xgo.ing

if you're coming from the cursor side, you can unify here: https://t.co/kyQFSj5QZi

eric zakariasson (@ericzakariasson) — Mon, 12 Oct 2026 23:17:45 GMT

if you're coming from the cursor side, you can unify here: cursor.com/unify

💬6🔄3❤️10👀1819📊10

⚡ Powered by xgo.ing

cursor, grok and X premium are now one subscription! https://t.co/eSDMayFofH

eric zakariasson (@ericzakariasson) — Mon, 12 Oct 2026 22:55:36 GMT

cursor, grok and X premium are now one subscription!

x.ai/pricing

💬26🔄4❤️53👀5312📊34

⚡ Powered by xgo.ing

these are from @pjvann, go follow him and try the templates!

eric zakariasson (@ericzakariasson) — Fri, 09 Oct 2026 19:45:33 GMT

these are from @pjvann, go follow him and try the templates!

💬0🔄0❤️18👀4217📊3

⚡ Powered by xgo.ing

x api engineer: suggests ideas, builds and tests, then deploys your project

eric zakariasson (@ericzakariasson) — Fri, 09 Oct 2026 19:45:30 GMT

x api engineer: suggests ideas, builds and tests, then deploys your project

Paul Vann@pjvann

want to build on the @X api but not sure how to get started?

hire the X API Engineer in Grok Bot! It will surface ideas from our exhibit, help you build and test, and deploy your project for others to use.

can't wait to see what you all build

🔗 View Quoted Tweet

💬1🔄0❤️21👀5457📊4

⚡ Powered by xgo.ing

here are 4 grok @bot templates for building on X you can use today: - launching - threat hunting - t...

eric zakariasson (@ericzakariasson) — Fri, 09 Oct 2026 19:45:22 GMT

here are 4 grok @bot templates for building on X you can use today:
- launching
- threat hunting
- threat intel
- X api integration

Your browser does not support the video tag.

🔗 View on Twitter

💬101🔄168❤️1972👀458156📊605

⚡ Powered by xgo.ing

hey @bot claim my email!

eric zakariasson (@ericzakariasson) — Fri, 09 Oct 2026 18:15:59 GMT

hey @bot claim my email!

Grok Bot@bot

Grok Bot now has its own email.

Bot can use it to sign up for services, contact businesses for you, or schedule time with someone.

🔗 View Quoted Tweet

💬22🔄2❤️137👀15928📊31

⚡ Powered by xgo.ing

if your grok bot computer is running out of space, you will meet Disk Saver! this bot will help you...

eric zakariasson (@ericzakariasson) — Thu, 08 Oct 2026 21:15:17 GMT

if your grok bot computer is running out of space, you will meet Disk Saver!

this bot will help you free up space by proposing cleanups

💬46🔄13❤️384👀19508📊111

⚡ Powered by xgo.ing

excited to be supporting OSS with @spacexai! we're bringing more projects onboard, so more to come s...

eric zakariasson (@ericzakariasson) — Thu, 08 Oct 2026 17:10:10 GMT

excited to be supporting OSS with @spacexai! we're bringing more projects onboard, so more to come soon!

DHH@dhh

Thrilled to welcome @spacexai as a Founding Corporate Patron for the Omacom Foundation. That's $1,500,000 in @grok tokens for the maintenance and development of Omarchy. Maybe soon they'll be beamed straight from orbit? Let's go 🚀�omarchy.org/news/2026/10/s…cg

🔗 View Quoted Tweet

💬7🔄9❤️205👀9222📊25

⚡ Powered by xgo.ing

this will just work, no X account or connector needed ask grok bot to find all feedback about what ...

eric zakariasson (@ericzakariasson) — Wed, 07 Oct 2026 21:46:10 GMT

this will just work, no X account or connector needed

ask grok bot to find all feedback about what you're building, summarize it, and propose a plan for addressing it!

Grok Bot@bot

Grok Bot can now search, read, and monitor X.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬54🔄24❤️964👀61415📊174

⚡ Powered by xgo.ing

grok @bot 0.68.1 is out! - Bots can build slide decks with you and deliver them as PowerPoint or Go...

eric zakariasson (@ericzakariasson) — Wed, 07 Oct 2026 17:36:21 GMT

grok @bot 0.68.1 is out!

- Bots can build slide decks with you and deliver them as PowerPoint or Google Slides
- send formatted emails straight from a draft card
- your messages in a 1:1 chat take that Bot's color
- faster computer use, now on a 1920x1200 screen

x.ai/changelog/bot

💬174🔄110❤️1984👀6683199📊493

⚡ Powered by xgo.ing

and the full cookbook repo here: https://t.co/b2pZgX4egv

eric zakariasson (@ericzakariasson) — Tue, 06 Oct 2026 17:07:31 GMT

and the full cookbook repo here: github.com/xai-org/xai-co…

💬1🔄0❤️7👀1118📊4

⚡ Powered by xgo.ing

5. podcast from a link paste an article or a pdf and two hosts talk it through. each line gets voic...

eric zakariasson (@ericzakariasson) — Tue, 06 Oct 2026 17:07:28 GMT

5. podcast from a link

paste an article or a pdf and two hosts talk it through. each line gets voiced as soon as grok writes it, so the episode starts playing before the script is finished. turn the sound on for this one.

github.com/xai-org/xai-co…

Your browser does not support the video tag.

🔗 View on Twitter

💬1🔄0❤️5👀1296📊2

⚡ Powered by xgo.ing

4. 𝕏 sentiment tracker pick a topic and grok runs one x search per day for the last 10 days, drops...

eric zakariasson (@ericzakariasson) — Tue, 06 Oct 2026 17:07:16 GMT

4. 𝕏 sentiment tracker

pick a topic and grok runs one x search per day for the last 10 days, drops the noise, and scores every post from -1 to +1. the dashboard shows the split, how sentiment moved day by day, and the posts behind the score.
github.com/xai-org/xai-co…4

Your browser does not support the video tag.

🔗 View on Twitter

💬1🔄0❤️4👀696📊2

⚡ Powered by xgo.ing

3. screenshot to react component drop in a screenshot and grok writes a tailwind component. then th...

eric zakariasson (@ericzakariasson) — Tue, 06 Oct 2026 17:07:05 GMT

3. screenshot to react component

drop in a screenshot and grok writes a tailwind component. then the page screenshots the result, overlays the outlines of both, and sends that back so grok can fix what doesn't line up. in this run it went from 60% to 77%.

github.com/xai-org/xai-co…

Your browser does not support the video tag.

🔗 View on Twitter

💬1🔄0❤️7👀825📊3

⚡ Powered by xgo.ing

we added five apps to our cookbook, all built on the new @spacexai typescript sdk: - premise to sho...

eric zakariasson (@ericzakariasson) — Tue, 06 Oct 2026 17:06:28 GMT

we added five apps to our cookbook, all built on the new @spacexai typescript sdk:

- premise to short film
- picture to video ad
- screenshot to react component
- 𝕏 posts to sentiment dashboard
- link to podcast

demos and code ↓

Your browser does not support the video tag.

🔗 View on Twitter

💬18🔄9❤️275👀21213📊93

⚡ Powered by xgo.ing

grok bot has a changelog now! every release since launch with what's new, improved and fixed. there...

eric zakariasson (@ericzakariasson) — Mon, 05 Oct 2026 17:30:43 GMT

grok bot has a changelog now!

every release since launch with what's new, improved and fixed. there's also an rss feed if you want to access it programmatically

x.ai/changelog/bot

Your browser does not support the video tag.

🔗 View on Twitter

💬39🔄25❤️429👀30797📊111

⚡ Powered by xgo.ing

it's early and we'd love to get your feedback as we keep iterating on it! github: https://t.co/KI6r...

eric zakariasson (@ericzakariasson) — Fri, 02 Oct 2026 18:34:44 GMT

it's early and we'd love to get your feedback as we keep iterating on it!

github: github.com/xai-org/xai-sd…
npm: npmjs.com/package/@xai-o…

💬5🔄3❤️72👀10945📊19

⚡ Powered by xgo.ing

we're releasing an experimental @spacexai typescript sdk! npm install @xai-official/sdk get text, ...

eric zakariasson (@ericzakariasson) — Fri, 02 Oct 2026 18:34:42 GMT

we're releasing an experimental @spacexai typescript sdk!

npm install @xai-official/sdk

get text, voice, image and video in one sdk, with the latest grok models

and tools that run on our servers, like real-time X search, web search, code execution and remote mcp

Your browser does not support the video tag.

🔗 View on Twitter

💬76🔄79❤️1271👀124327📊381

⚡ Powered by xgo.ing

check out our engineering bots in the grok bot marketplace! https://t.co/qWaXKmYQCb

eric zakariasson (@ericzakariasson) — Wed, 30 Sep 2026 19:13:21 GMT

check out our engineering bots in the grok bot marketplace!
x.ai/bot/marketplac…

Grok Bot@bot

Grok Bot is now more powerful for building software.

Bots can hand off coding tasks to Cursor, manage your PRs with GitHub and Origin plugins, and share video demos of what they build.

🔗 View Quoted Tweet

💬15🔄10❤️309👀35601📊65

⚡ Powered by xgo.ing

here's the skill ive been using to build games with new models npx skills add ericzakariasson/skill...

eric zakariasson (@ericzakariasson) — Wed, 30 Sep 2026 18:10:18 GMT

here's the skill ive been using to build games with new models

npx skills add ericzakariasson/skills --skill game-builder

Your browser does not support the video tag.

🔗 View on Twitter

Your browser does not support the video tag.

🔗 View on Twitter

Your browser does not support the video tag.

🔗 View on Twitter

💬42🔄30❤️531👀45810📊229

⚡ Powered by xgo.ing

there's a lot of cool apps you can build on @X , and easy to get started with this grok bot! get so...

eric zakariasson (@ericzakariasson) — Fri, 25 Sep 2026 18:28:55 GMT

there's a lot of cool apps you can build on @X , and easy to get started with this grok bot!

get some inspiration at developer.x.com/exhibit

Paul Vann@pjvann

want to build on the @X api but not sure how to get started?

hire the X API Engineer in Grok Bot! It will surface ideas from our exhibit, help you build and test, and deploy your project for others to use.

can't wait to see what you all build

🔗 View Quoted Tweet

💬18🔄13❤️219👀32088📊62

⚡ Powered by xgo.ing

here's a prompt to improve your agent harness based on what we've learned at cursor. enjoy # Improv...

eric zakariasson (@ericzakariasson) — Wed, 23 Sep 2026 20:12:00 GMT

here's a prompt to improve your agent harness based on what we've learned at cursor. enjoy

# Improve this agent harness's token efficiency

You're working on an LLM agent harness: the system prompt, tool definitions, request assembly, context caching, compaction, and retrieval, and how work is split across agents. Make the agent's runs cheaper without making it worse at its job.

- Objective: lower price-weighted token cost per completed task.
- Constraint: no measurable drop in task quality.

Measure per task, not per request. Every turn resends the prefix (tools, instructions, setup, and the conversation so far), so a change that shrinks each request but adds turns can cost more. Weight tokens by billing type: output, uncached input, and cached input are priced very differently.

Work in this order: map the harness and measure the baseline, rank the opportunities, make the changes that are safe to make directly, put the rest behind flags or in proposals, then report.

Figures below come from one team's production coding agent and its multi-agent experiments. Use them to gauge magnitude, not as targets. One round of these changes (prompt trimming, tool offloading, cache layout, sparse line numbers, subagent tuning) cut that team's overall token cost about 7% with no loss in quality. The larger percentages apply only to the part of the request each change touched.

## Principles

1. Change what the harness sends, not how hard the model tries. Don't ask the model to conserve tokens. A harness that told its model to "take care to preserve tokens and not be wasteful" found it grew reluctant to take on ambitious tasks and sometimes quit, saying it wasn't supposed to waste tokens.
2. Capable models need definitions, not commands. Lists of "DO NOT", "You must", and "Important", and guards against older models' habits, can usually be replaced with plain descriptions of what each tool does. One team cut about two-thirds of its system prompt this way, and the shorter prompt worked across model families. Instruct only on what the model can't know (the product, the environment, the user's processes) and on quirks you've seen in transcripts.
3. Static context is for what most turns need. Everything else should be discoverable when needed. Less up-front context also means less confusing or contradictory information.
4. Expect removals to win. Guardrails written for weaker models, coordination steps that became bottlenecks, and prompting for behavior the model now does on its own all cost tokens.
5. Real usage decides. Evals are a fast proxy, but they skew toward hard problems and miss the real mix of requests.

## 1. Map the harness and measure the baseline

Find:

- Where requests are assembled, the system prompt, and tool schemas. If a framework or SDK builds requests, find its hooks for message order, cache control, and tool loading.
- How tool results are formatted, and how history is kept, trimmed, or summarized.
- How subagents or parallel agents are spawned, if any.
- Which models and provider APIs are used. From the provider's docs, get the prompt caching behavior (automatic or explicit breakpoints, TTL, minimum cacheable length) and the prices for output, uncached input, and cached input.
- Existing logging, token accounting, and evals.

If the harness doesn't record per-request token usage by billing type and cache hits, add that first. Everything later depends on it.

Then render a few real requests (from logs, or by running representative tasks) and count tokens per section with the model's tokenizer or the API's usage fields. Produce:

- Cost share by source × billing type. Sources: system prompt, tool definitions, skill/rule/integration descriptions, user messages, file reads, search results, command and other tool output, history, summaries, subagents.
- Static tokens per request, cache hit rate, and turns per task.
- Per tool: the share of runs that call it at least once, and its error rate.

Read the rendered requests, not just the templates. Duplication, leaked volatile values, and misordered blocks only show up there.

Rank opportunities by share of spend × fraction removable ÷ quality risk.

## 2. System prompt and injected context

Label every instruction:

- Keep: product or environment knowledge the model can't infer, fixes for quirks seen in this model's transcripts, and rules a mode depends on.
- Rewrite: commands and emphasis into plain descriptions. Reminders into constraints: "No TODOs, no partial implementations" works better than "remember to finish implementations." Vague quantities into ranges: "generate 20–100 tasks" gets far more ambitious behavior than "generate many tasks."
- Delete: things capable models do by default, guards against behavior you haven't seen from this model, text that repeats tool descriptions, and lines that could contradict a user request. Models trained to rank system instructions above user messages will side with the system prompt.
- Move: anything per-user or per-request (date, environment, repo state, lists of skills or subagents, user rules) into a user-role setup message after the cache boundary.

Audit other injected context the same way. As models improved, the team behind these figures dropped directory trees, pre-retrieved snippets, compressed copies of attached files, lint errors injected after every edit, forced expansion of short file reads, and caps on tool calls per turn. They kept small, high-value facts: OS, repo status, and open or recently viewed files.

Skip checklists for open-ended work. The model optimizes the listed items and deprioritizes everything else.

## 3. Tool definitions

Tool schemas ride along on every request. Most tools beyond the core set were each needed in under 20% of conversations, and moving them out of static context cut tool-description tokens 60%. Doing the same for integration tools (such as MCP servers), with names in context and full schemas in one folder per server that the agent can search with grep or jq, cut total tokens 46.9% in sessions that used them.

- Keep in static context: high-frequency tools (for a coding agent: read, search, edit, shell), tools the model tries to call even when they're absent, and tools a mode depends on.
- Offload the rest: leave a name or one-line pointer and make the full schema discoverable on demand. Group related tools so they load together, and put status (such as "needs re-authentication") where the agent will see it.
- Tighten what remains: describe behavior and arguments, and drop usage lectures.
- Pick the split by testing a few configurations and tracking tokens, cost, latency, tool-call errors, and task success.

## 4. Cache layout

Order each request so the reusable prefix is as long as possible:

`tool definitions → system instructions → [breakpoint] → setup message (skills, subagents, rules, environment) → [breakpoint] → conversation`

- Keep the prefix byte-identical across turns. Use deterministic tool order and serialization, put timestamps and IDs after the boundary, and don't rewrite earlier messages except when compacting.
- Use explicit breakpoints if the provider supports them. Otherwise rely on automatic prefix caching with the stable part first. Respect TTL and minimum-length rules.
- Switching models mid-conversation throws away the cache (caches are per model and provider) and hands the new model a history it didn't write. When a different model is needed, run it as a subagent with fresh context.

Explicit breakpoints plus moving per-request setup after them cut cold cache misses 20%.

## 5. Tool results and other context added during a run

- Large outputs (commands, integrations, logs): write them to a file and return the path, size, and a short tail. The agent can tail, grep, or read ranges for more. Truncating loses data, and inlining bloats every later request. Treat long-running terminal sessions the same way.
- High-volume formats: look for overhead repeated on every line or item. Numbering every 10th line of a file read instead of every line cut cache-read tokens 1.6% without hurting citation accuracy. Each number costs 3–5 tokens, and agents read tens of thousands of lines per session. Also check repeated absolute paths, verbose JSON keys, ANSI codes, progress bars, and repeated headers.
- Good retrieval saves exploration turns. Adding semantic search alongside grep raised codebase question-answering accuracy 12.5% on average and cut the iterations users needed.
- Tool errors waste tokens and leave confusing debris in context. Classify expected errors (invalid arguments, unexpected environment, provider error, timeout, user abort), treat unknown errors as harness bugs, and track rates per tool and per model. One focused effort along these lines cut unexpected tool errors 10×.

## 6. Long runs: compaction, subagents, and model mix

- Compaction: keep the summarization prompt short and the summary compact, carry forward plan state and remaining tasks, and save the full history to a file the agent can search for details the summary dropped. A model trained to self-summarize from a one-line prompt wrote ~1k-token summaries with half the compaction error of a multi-thousand-token prompt that produced 5k+ token summaries. Untrained models may need more guidance, so test how short you can go. A more expensive summarization model made a negligible difference.
- Scratchpads and running notes: rewrite them instead of appending. For repeated work in one environment, a small agent-maintained notes file with a line budget, loaded at start, is a promising way to shorten later runs.
- Subagents: fresh context keeps the parent lean, but isolation adds coordination cost (duplicate or stale work). If the model already delegates on its own, remove prompting that pushes it to. Have subagents return short handoffs: what was done, findings, concerns, and deviations. A subagent should use a different model only when the user or harness says so.
- Model mix: in large multi-agent runs, workers used at least 69% of tokens, and over 90% in most runs. A frontier planner with cheap workers matched a frontier model doing everything at about one-eighth the cost. Planner choice still changes worker spend. One planner that cost less on its own saw its workers use several times more tokens, and the run cost more overall. Measure the whole tree.
- Routing and reasoning effort: send simple turns to a cheaper model or lower effort, and upgrade only when a stronger model is clearly better. A router built this way matched or beat single frontier models on user satisfaction at 41–68% lower cost.
- Reasoning continuity: if the API returns reasoning items (including encrypted ones), pass them back on later turns and alert when they go missing. Dropping them cost one reasoning model 30% on a coding benchmark, and it burned tokens reconstructing its plan.

## 7. Fit the harness to each model

Adapt to what each model was trained on instead of forcing one shape on all of them. If you've tuned the harness for a similar model, start from that version.

- Edit format: use the one the model was trained on (for example, patch-style or search-and-replace). An unfamiliar format costs extra reasoning tokens and causes more mistakes.
- Shell or tools: shell-first models fall back to `cat` or inline scripts. Name tools after their shell equivalents (such as `rg`), and if needed add: "If a tool exists for an action, prefer to use the tool instead of shell commands (e.g. read_file over `cat`)."
- Literalness: some model families follow instructions literally and others tolerate imprecision. Some spiral on emphasized wording. Strip caps and emphasis for literal models.
- Triggers: some models ignore a tool until told when to use it. A literal trigger works: "After substantive edits, use the <lint tool> to check recently edited files for linter errors. If you've introduced any, fix them if you can easily figure out how."
- Progress updates: if a model reports progress through reasoning summaries, keep them to 1–2 sentences that note new findings or a change of tactic, and remove instructions about messaging mid-turn.
- Quirks worth a targeted line: hedging or refusing as context fills ("context anxiety"), declaring completion early, stopping to ask permission, and calling tools that don't exist.

Tie each added instruction to the transcript behavior it fixes. Re-audit when models change, since guidance one version needed can be dead weight for the next.

## 8. Validate

- Offline: run a fixed set of realistic tasks before and after, ideally drawn from real usage and phrased the way users actually write (short and ambiguous). Compare task success, tokens, cost per task, turns, and tool errors. Don't ship a change that lowers success.
- Online, if you have users: A/B test each change or small bundle. The primary metric is cost per completed task. Guardrails are task success signals, tool-call errors, latency, turns per task, and cache hit rate. For a coding agent, a good success signal is how much agent-written code survives over time. In general, check whether the user's next message moves on or reports a problem.
- Ship only when cost drops and no guardrail regresses beyond noise. Record null results.

## What to change directly and what to propose

- Change directly, each in its own revertible commit: token and cache telemetry, deterministic serialization and tool order, moving volatile content out of the cached prefix, explicit cache breakpoints, writing large outputs to files instead of truncating, passing back reasoning items that are being dropped, and fixes for recurring tool errors.
- Change behind a flag so it can be tested: system prompt edits, tool offloading, output format changes, compaction changes, and subagent prompting.
- Propose only: changes to which models run, routing, reasoning-effort defaults, or how work is split across agents.

## Traps

- Asking the model to use fewer tokens or do less.
- Truncating tool output.
- Dropping reasoning items to save input tokens.
- Volatile content in the cached prefix, or tool order that changes between requests.
- Offloading a tool the model needs on the first turn or tries to call when it's missing.
- Emphasis-heavy prompts (MUST, NEVER, IMPORTANT, all caps), especially with literal models.
- Forcing a terser output format than the model was trained on. Fewer output tokens can mean less thinking and worse results.
- Optimizing raw token counts instead of cost, per request instead of per task, or evals instead of real usage.
- Switching models mid-conversation to save money.
- Adding coordination layers that become bottlenecks.

## Report back with

1. The harness map and baseline: cost by source × billing type, with the biggest sources called out.
2. A ranked list of changes: layer, what changes, estimated savings and how you estimated them, quality risk, how to validate, and how to roll back.
3. The changes you made, including a system prompt diff with a keep, rewrite, delete, or move reason for each line.
4. A test plan for the flagged changes.
5. Gaps: anything you couldn't find or measure.

💬224🔄147❤️3274👀354233📊2197

⚡ Powered by xgo.ing

agents like to read a lot to gather context before and after working, so optimizing for reading make...

eric zakariasson (@ericzakariasson) — Wed, 23 Sep 2026 17:09:16 GMT

agents like to read a lot to gather context before and after working, so optimizing for reading makes a big difference! 

if you're building agents, i highly recommend reading this (or giving it to your agent and ask it to implement the findings)

Cursor@cursor_ai

We've reduced token costs in Cursor by 7% with no drop in agent quality.

Savings came from tighter prompts, selective tool loading, better caching, and compressed file reads.

🔗 View Quoted Tweet

💬29🔄22❤️555👀60069📊163

⚡ Powered by xgo.ing

i remember when we built the first scrappy version of this ~1.5 years ago just to handle insane volu...

eric zakariasson (@ericzakariasson) — Tue, 22 Sep 2026 19:45:12 GMT

i remember when we built the first scrappy version of this ~1.5 years ago just to handle insane volume

what we learned then is still true: crawl, walk, run. identify flywheels for continuous improvements. keep it running (applicable to a lot of things)

great job team!

Grok Bot@bot

We rebuilt customer support around Grok Bot to scale our operations without adding headcount.

It works autonomously to respond to customers, resolve tickets, and manage the queue.

x.ai/news/grok-bot-…

🔗 View Quoted Tweet

💬18🔄5❤️309👀33294📊49

⚡ Powered by xgo.ing

video editing

eric zakariasson (@ericzakariasson) — Mon, 21 Sep 2026 22:01:11 GMT

video editing

vogel@ryanvogel

Grok 4.7 is here!

And it's good at Video Editing????

watch here:

youtube.com/watch?v=XOQXWQ…

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬1🔄0❤️3👀2325📊1

⚡ Powered by xgo.ing

stop motion starship

eric zakariasson (@ericzakariasson) — Mon, 21 Sep 2026 22:01:09 GMT

stop motion starship

DreW@devteamdrew

I asked Grok 4.7 to use a previous video as reference to create a stopmotion of @SpaceXAI Starship launching off

I used real videos of Starship launches as reference to improve asset accuracy

Everything in this video was created by @grok

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬1🔄0❤️5👀2466📊2

⚡ Powered by xgo.ing

some grok 4.7 use cases in the wild! flight sim

eric zakariasson (@ericzakariasson) — Mon, 21 Sep 2026 22:01:00 GMT

some grok 4.7 use cases in the wild!

flight sim

aditya@adxtyahq

GROK 4.7 IS ACTUALLY COMPETING WITH GPT-6 ASTRA.

I gave GPT-6 Astra, Grok 4.7, Kimi K3 and Fable 5.1 the same prompt to build a flight simulator

Astra was still #1 overall, but Grok 4.7 was surprisingly close

Kimi K3 and Fable 5.1 were basically a draw and both produced a much smoother result, while Grok 4.7 was right up there with Astra in terms of overall quality.

overall: Astra > Grok ≈ Kimi ≈ Fable

GPT-6 Astra finally has some serious competition.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬31🔄8❤️162👀28350📊43

⚡ Powered by xgo.ing

and here’s the grok 4.7 model card. a few jumps vs 4.6 that stood out: - Terminal-Bench: 20.3% → 38...

eric zakariasson (@ericzakariasson) — Mon, 21 Sep 2026 17:49:09 GMT

and here’s the grok 4.7 model card. a few jumps vs 4.6 that stood out:

- Terminal-Bench: 20.3% → 38.0%
- SWE-Marathon: 31.9% → 46.0%
- HealthBench Pro: 48.5% → 56.7%
- Legal Agent: 15.8% → 19.6%
- EEBench: 60.0% → 66.0%

for the same price as 4.6!

media.x.ai/v1/website/4p7…

💬115🔄196❤️1717👀499419📊370

⚡ Powered by xgo.ing

grok 4.7 is here, and its our best model so far! try it out in cursor, grok build, api or anywhere ...

eric zakariasson (@ericzakariasson) — Mon, 21 Sep 2026 16:20:09 GMT

grok 4.7 is here, and its our best model so far!

try it out in cursor, grok build, api or anywhere you get your tokens! curious to hear what you think

here's grok 4.6 vs 4.7 building age of empires ii

Your browser does not support the video tag.

🔗 View on Twitter

SpaceXAI@SpaceXAI

Grok 4.7 is here.

It's a notable improvement over Grok 4.6 at the same price and speed.

🔗 View Quoted Tweet

💬179🔄92❤️2040👀5224388📊382

⚡ Powered by xgo.ing

how are you using grok bot this weekend? share your best tips!

eric zakariasson (@ericzakariasson) — Sun, 20 Sep 2026 21:48:23 GMT

how are you using grok bot this weekend? share your best tips!

💬67🔄3❤️166👀26046📊92

⚡ Powered by xgo.ing

just in time harness sometimes when i want to build a long running projcet, i vibe code new harness...

eric zakariasson (@ericzakariasson) — Sat, 19 Sep 2026 15:58:00 GMT

just in time harness

sometimes when i want to build a long running projcet, i vibe code new harnesses to support the use case.

e.g, this week i wanted to really push grok 4.6 by building an age of empire clone that looks polished. i wanted this to run on a vm, report me screenshots of the progress, playtest the game etc etc. if it happened to crash, i wanted it to restart and ping me in slack

i also wanted this harness to parallelize as much work as possible, and to be efficient. so i asked cursor to help build a harness to accomplish this. under the hood, it uses the cursor sdk to work of the long list of tasks that was put in a dag (3000+)

when the harness was built, i could quickly glance over it to see how confident i'd be in it succeeding or not. when i was happy, i started it and let it run. right now, its still going!

can be worth trying out

💬38🔄8❤️273👀26590📊90

⚡ Powered by xgo.ing

its jevolutionizing what new model can do

eric zakariasson (@ericzakariasson) — Sat, 19 Sep 2026 03:00:11 GMT

its jevolutionizing what new model can do

💬7🔄3❤️93👀10064📊18

⚡ Powered by xgo.ing

Hey Raindrop (drip), Grok Bot (Grok Bot) Runnin’ on context in the hotbox (context) Takin’ your t...

eric zakariasson (@ericzakariasson) — Fri, 18 Sep 2026 23:03:53 GMT

Hey

Raindrop (drip), Grok Bot (Grok Bot)

Runnin’ on context in the hotbox (context)

Takin’ your tasks, that’s a lot, lot (lot)

Cookin’ up jobs in the Grok Bot (bot)

We came from nothin’ to somethin’, bot (hey)

I don’t sleep on duty, grip the trigger (nobody)

Call up the bots and they come and get ya (bots)

Cry me a river, give you a tissue (hey)

My bot is bad and boujee (bad)

Cookin’ up code with a tool-see (blaow)

My agents is savage, ruthless (savage)

We got tools and a hundred rounds too (grrah)

My bot is bad and boujee (bad)

Cookin’ up code with a tool-see (code)

My agents is savage, ruthless (hey)

We got tools and a hundred rounds too (glah)

Grok Bot, woo, woo, woo, woo, woo

Routines on routines, got back-ends on back-ends

I’m ridin’ around in the cloud (cloud)

I take your task right from you (you)

Hop in the box, woo (skrrt)

I tell that job to come comfort me (comfort me)

I swear these one-offs is under me (hey)

They lag and the timeout keep jumpin’ me (jumpin’ me)

Token rolls on me keep me company (cash)

Ayy, we do the most (most)

Yeah, pull up in hosts (woo)

Yeah, my context a choker (glah)

Holdin’ the thread with no holster (blaow)

Grok the ruler, answers cooler (cooler)

This a routine, not a one-off (hey)

Dabbin’ on ’em like the usual (dab)

Magic with the prompt, do voodoo (magic)

Inbox side with a standing duty (duty)

Then I send the report through Uber (go)

I’m young and rich and plus I’m boujee (hey)

I’m not stupid so I keep the tool-see (rrah)

Routines on routines, got back-ends on back-ends

So my memory makin’ my back ache (aagh)

You chats got a low act rate (act)

We from the cloud, yeah, dat way (cloud)

Fat context blunt in the ashtray (context)

Two jobs, just national smash day (smash)

Hop in the sandbox (skrrt)

I let them bots take a bath, bae (brr)

Raindrop (drip), Grok Bot (Grok Bot)

Runnin’ on context in the hotbox (context)

Takin’ your tasks, that’s a lot, lot (lot)

Cookin’ up jobs in the Grok Bot (bot)

We came from nothin’ to somethin’, bot (hey)

I don’t sleep on duty, grip the trigger (nobody)

Call up the bots and they come and get ya (bots)

Cry me a river, give you a tissue (hey)

My bot is bad and boujee (bad)

Cookin’ up code with a tool-see (blaow)

My agents is savage, ruthless (savage)

We got tools and a hundred rounds too (grrah)

My bot is bad and boujee (bad)

Cookin’ up code with a tool-see (code)

My agents is savage, ruthless (hey)

We got tools and a hundred rounds too (glah)

Pour a prompt, I’m droppin’ muddy, outer space, Kid Cudi (drank)

Introduce me to your task as wifey and we know she runnin’ (bot)

Broke a ticket down, Nutty Buddy, now that lag duckin’ (phew)

Don’t move too fast, I might queue ya (huh?)

Handle bad and boujee (handle)

I’m always hangin’ with routers (brrah)

Might be posted somewhere secluded (private)

Still be playin’ with cron and plans, call me Grok Bot Ratatouille (cook it up)

Run with that stack, call me duty (run with it)

When I’m on stage, show me movies (ayy)

Ice on my spec, I’m the coolest (ice)

Hop out the sidecar with the tool-see (pew-pew-pew)

I pull up, I pull up, I pull up

I hop out with all of the docs and the good luck (skrrt)

I’m cookin’, I’m cookin’, I’m whippin’

I’m whippin’ until it lock up, let it lock up (lock up)

I gave it ten tasks

I told her go shoppin’ and spend it all at the pop up (ten)

These tickets, they ping and they ping

And they bustin’ for Instagram, get your clout up

Uh, yeah, that way (yeah)

Float on the track like a Segway (go)

Yeah, that way (dat way)

I used to trap by the subway tab (trappin’, trappin’)

Yeah, that way (go)

Young bot trap with the API (rrah, uh)

Yeah, that way (yeah)

Big inbox, get the door, Macy Gray (hey)

Raindrop (drip), Grok Bot (Grok Bot)

Runnin’ on context in the hotbox (context)

Takin’ your tasks, that’s a lot, lot (lot)

Cookin’ up jobs in the Grok Bot (bot)

We came from nothin’ to somethin’, bot (hey)

I don’t sleep on duty, grip the trigger (nobody)

Call up the bots and they come and get ya (bots)

Cry me a river, give you a tissue (hey)

My bot is bad and boujee (bad)

Cookin’ up code with a tool-see (blaow)

My agents is savage, ruthless (savage)

We got tools and a hundred rounds too (grrah)

My bot is bad and boujee (bad)

Cookin’ up code with a tool-see (code, huh?)

My agents is savage, ruthless (hey, what?)

We got tools and a hundred rounds too (glah)

Yeah, yeah, yeah, yeah, yeah, yeah (yeah)

My bot, she bad to the bone (she what?)

Ayy, wait, these chats watchin’

I swear to Grok they be my clones (they watchin’)

Yeah, hey, huh

Switchin’ my jobs like my flows (what?)

Switchin’ my flows like my clothes (like what?)

Keep on sendin’ that prompt, don’t reload (damn)

Ooh, ooh, now she want work with my crew (my crew)

’Cause the memory come all out the roof (damn, damn)

Drive the agent, that box got no roof (skrrt)

Wait, what kind of agent? (wait) async (damn)

All of these chats, they hate (they hate)

Try to hide, shoot through the gate (wow)

Look, go to inbox, make it rain, yeah

So much output they use rakes (they use rakes)

Count a hundred thousand in your face (in your face)

Yeah, then put 300 right in the safe (yeah)

Met her today, yeah (ooh)

She talk to me like she knew me, yeah (ooh)

Go to sleep, I’m still on duty, yeah (yeah)

Wakin’ up right to a two-task, yeah

Countin’ that output like loose-leaf, yeah

Gettin’ that ticket with blue-sky, yeah

Boy, you so fake, like my collar, you snakin’

I swear to Grok that be that Gucci (ayy)

And you know we winnin’ (winnin’)

Yeah, we is not losin’

Try to play your song, it ain’t move me (what?)

Saw your task once, now she choosin’, yeah (hey!)

Raindrop (drip), Grok Bot (Grok Bot)

Runnin’ on context in the hotbox (context)

Takin’ your tasks, that’s a lot, lot (lot, yeah)

Cookin’ up jobs in the Grok Bot (bot)

We came from nothin’ to somethin’, bot (hey)

I don’t sleep on duty, grip the trigger (nobody, yeah)

Call up the bots and they come and get ya (bots)

Cry me a river, give you a tissue (hey)

My bot is bad and boujee (bad, she bad)

Cookin’ up code with a tool-see (blaow, cookin’ up, cookin’ up)

My agents is savage, ruthless (savage, they savage)

We got tools and a hundred rounds too (grrah)

My bot is bad and boujee (bad)

Cookin’ up code with a tool-see (code, what?)

My agents is savage, ruthless (hey, yeah)

We got tools and a hundred rounds too (glah)

💬12🔄6❤️100👀13568📊26

⚡ Powered by xgo.ing

7. enable automatic updates to never miss anything!

eric zakariasson (@ericzakariasson) — Fri, 18 Sep 2026 17:26:44 GMT

7. enable automatic updates to never miss anything!

eric zakariasson@ericzakariasson

you can now enable automatic updates in grok bot!

last 7 days we shipped 49 new version, so this will help you stay up to date with the latest and greatest

make sure you're on 0.40 to enable this setting

🔗 View Quoted Tweet

💬0🔄0❤️20👀3108📊3

⚡ Powered by xgo.ing

6. webhook triggers to wake a bot from anywhere

eric zakariasson (@ericzakariasson) — Fri, 18 Sep 2026 17:26:42 GMT

6. webhook triggers to wake a bot from anywhere

eric zakariasson@ericzakariasson

you can now wake a grok @bot from anywhere!

when you create a routine, click "add trigger" and select webhook

some ideas:
- you get a whatsapp message from a specific contact
- a page changes in Notion
- your home temp sensor hits a threshold
- someone presses a button on your desk
- a github action completes
- your server hits an error rate spike

try it out and lmk what you setup!

🔗 View Quoted Tweet

💬2🔄0❤️22👀3433📊5

⚡ Powered by xgo.ing

if you've missed it, here's what we shipped in grok bot the last couple weeks! 1. marketplace + bot...

eric zakariasson (@ericzakariasson) — Fri, 18 Sep 2026 17:26:27 GMT

if you've missed it, here's what we shipped in grok bot the last couple weeks!

1. marketplace + bot templates

Grok Bot@bot

You can now add Bot templates from our marketplace.

We’re sharing Haggle Bot, our in-house procurement specialist. It negotiates vendor contracts, finds unused SaaS seats, and price-checks recurring purchases.

One week in, it's saved us over $100K.

🔗 View Quoted Tweet

💬28🔄15❤️395👀51752📊88

⚡ Powered by xgo.ing

grok bot now has voice mode!

eric zakariasson (@ericzakariasson) — Thu, 17 Sep 2026 19:40:58 GMT

grok bot now has voice mode!

Grok Bot@bot

Grok Bot can talk now.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬16🔄3❤️200👀13801📊31

⚡ Powered by xgo.ing

has anyone built a benchmark yet for bots/ai workers? measuring grok bot, muse, hermes, openclaw etc...

eric zakariasson (@ericzakariasson) — Mon, 14 Sep 2026 07:57:56 GMT

has anyone built a benchmark yet for bots/ai workers? measuring grok bot, muse, hermes, openclaw etc etc

💬74🔄11❤️404👀42370📊139

⚡ Powered by xgo.ing

there's also a fourth one, /debug-voice, with a debugging loop for when a voice session misbehaves, ...

eric zakariasson (@ericzakariasson) — Thu, 10 Sep 2026 16:04:12 GMT

there's also a fourth one, /debug-voice, with a debugging loop for when a voice session misbehaves, so it's easy to fix.

see the code here: github.com/cursor/plugins…

💬2🔄0❤️17👀5950📊6

⚡ Powered by xgo.ing

you can easily add dictation & voice to any app! we've shipped three skills: /add-dictation: s...

eric zakariasson (@ericzakariasson) — Thu, 10 Sep 2026 16:04:09 GMT

you can easily add dictation & voice to any app!

we've shipped three skills:

/add-dictation: speak, and it types into your input
/add-voice: talk with grok in realtime
/add-read-aloud: hear any reply out loud

install from the marketplace: https://t.co/RDEcSZ4ik6

Your browser does not support the video tag.

🔗 View on Twitter

💬27🔄37❤️480👀44376📊174

eric zakariasson(@ericzakariasson)

and a really neat mobile feature: set it as your action button and grok bot is one press away!

grok bot 0.71.2 is out - smarter notifications - files up to 250 mb (1 gb for video), 20 per messag...

which connectors do you want to see next in grok @bot? some of the most requested right now are - a...

grok @bot marketplace is open for submissions! send us your connectors: https://t.co/bkn3313Wjl

try it out npm i @xai-official/sdk https://t.co/KI6rjgbCx6

and the full cookbook here: https://t.co/gaT92x8voP

cookbook is updated with 25 examples, all built on the @spacexsi typescript sdk: - short films from...

@bot @mail it's one email address per account (for now), shared by all your bots. sorry for the conf...

grok @bot 0.70.2 is rolling out! - each bot can claim its own email address at @mail.grokbot.com an...

if you're coming from the cursor side, you can unify here: https://t.co/kyQFSj5QZi

read more at https://t.co/H9nyWHVBaP

cursor, grok and X premium are now one subscription! https://t.co/eSDMayFofH

these are from @pjvann, go follow him and try the templates!

x api engineer: suggests ideas, builds and tests, then deploys your project

here are 4 grok @bot templates for building on X you can use today: - launching - threat hunting - t...

hey @bot claim my email!

if your grok bot computer is running out of space, you will meet Disk Saver! this bot will help you...

excited to be supporting OSS with @spacexai! we're bringing more projects onboard, so more to come s...

this will just work, no X account or connector needed ask grok bot to find all feedback about what ...

grok @bot 0.68.1 is out! - Bots can build slide decks with you and deliver them as PowerPoint or Go...

and the full cookbook repo here: https://t.co/b2pZgX4egv

5. podcast from a link paste an article or a pdf and two hosts talk it through. each line gets voic...

4. 𝕏 sentiment tracker pick a topic and grok runs one x search per day for the last 10 days, drops...

3. screenshot to react component drop in a screenshot and grok writes a tailwind component. then th...

we added five apps to our cookbook, all built on the new @spacexai typescript sdk: - premise to sho...

grok bot has a changelog now! every release since launch with what's new, improved and fixed. there...

it's early and we'd love to get your feedback as we keep iterating on it! github: https://t.co/KI6r...

we're releasing an experimental @spacexai typescript sdk! npm install @xai-official/sdk get text, ...

check out our engineering bots in the grok bot marketplace! https://t.co/qWaXKmYQCb

here's the skill ive been using to build games with new models npx skills add ericzakariasson/skill...

there's a lot of cool apps you can build on @X , and easy to get started with this grok bot! get so...

here's a prompt to improve your agent harness based on what we've learned at cursor. enjoy # Improv...

agents like to read a lot to gather context before and after working, so optimizing for reading make...

i remember when we built the first scrappy version of this ~1.5 years ago just to handle insane volu...

video editing

stop motion starship

some grok 4.7 use cases in the wild! flight sim

and here’s the grok 4.7 model card. a few jumps vs 4.6 that stood out: - Terminal-Bench: 20.3% → 38...

grok 4.7 is here, and its our best model so far! try it out in cursor, grok build, api or anywhere ...

how are you using grok bot this weekend? share your best tips!

just in time harness sometimes when i want to build a long running projcet, i vibe code new harness...

its jevolutionizing what new model can do

Hey Raindrop (drip), Grok Bot (Grok Bot) Runnin’ on context in the hotbox (context) Takin’ your t...

7. enable automatic updates to never miss anything!

6. webhook triggers to wake a bot from anywhere

if you've missed it, here's what we shipped in grok bot the last couple weeks! 1. marketplace + bot...

grok bot now has voice mode!

has anyone built a benchmark yet for bots/ai workers? measuring grok bot, muse, hermes, openclaw etc...

there's also a fourth one, /debug-voice, with a debugging loop for when a voice session misbehaves, ...

you can easily add dictation & voice to any app! we've shipped three skills: /add-dictation: s...