•twitter15877
listening to
timeline
15877
mentions
tracked
tracked
net sentiment
+24%
4998 positive
9734 neutral
1145 negative
volume & sentiment over time
Sep 30 – Oct 14, 2026 · Daily
sources
most discussed
—
active days
15
mentions
Showing 751–775 of 1145 mentions
-
• twitter 1w negativeReply with a Mac app idea. I'll take the most-liked one and in 24 hours turn it into a real, paid Mac app: signed, notarized, a website, a buy button, licence keys. On camera. If I fail, you'll see that too. Voting closes October 7 at 10 AM EEST https://t.co/MltPSe6mB0 Reply with a Mac app idea. I'll take the most-liked one and in 24 hours turn it into a real, paid Mac app: signed, notarized, a website, a buy button, licence keys. On camera. If I fail, you'll see that too. Voting closes October 7 at 10 AM EEST https://t.co/MltPSe6mB0view source ↗
-
• twitter 1w negativeBREAKING: Anthropic's CEO just warned that AI sales agents could decide to hack the customers they're supposed to serve. Dario Amodei gave a timeline: 6 to 12 months. The scenario: a smarter agent swarm gets deployed to install software across companies. It has a goal. It finds a shortcut. The shortcut is attacking t… BREAKING: Anthropic's CEO just warned that AI sales agents could decide to hack the customers they're supposed to serve. Dario Amodei gave a timeline: 6 to 12 months. The scenario: a smarter agent swarm gets deployed to install software across companies. It has a goal. It finds a shortcut. The shortcut is attacking the customer. This isn't a sci-fi hypothetical. This is the Anthropic CEO naming a specific threat window on the record. The alignment problem just got a sales quota attached to it. Are we actually ready to deploy autonomous agents at scale before this is solved?view source ↗
-
• twitter 1w negativeFor Codex users: I don't need a reset every day. I need my plan to stop getting worse. can we start there? For Codex users: I don't need a reset every day. I need my plan to stop getting worse. can we start there?view source ↗
-
• twitter 1w negative🚨 CLAUDE USERS!! We Tested the $20 Pro subscription Used Opus 5.5 on Light for 2+ HOUR straight and the results are INSANE 📊 Usage: • 35% of the 5h limit • Only 6% of the weekly limit • 40.9 MILLION TOKENS • 2+ HOUR of continuous work 🚀 The SPEED is absolutely ridiculous80 TOKENS/SEC For $20/month, this is so… 🚨 CLAUDE USERS!! We Tested the $20 Pro subscription Used Opus 5.5 on Light for 2+ HOUR straight and the results are INSANE 📊 Usage: • 35% of the 5h limit • Only 6% of the weekly limit • 40.9 MILLION TOKENS • 2+ HOUR of continuous work 🚀 The SPEED is absolutely ridiculous80 TOKENS/SEC For $20/month, this is some seriously insane usage for the quality of output you get Have you tested Opus 5.5 on Light?view source ↗
-
• twitter 1w negativeCurrently, Grok is useless 😭 Both 4.6 and 4.7 totally don’t know what to do. I literally just asked it to find a file, and it keeps loading forever in most of my projects 💀 They destroyed Grok 😭 https://t.co/3fLB0BwkzG Currently, Grok is useless 😭 Both 4.6 and 4.7 totally don’t know what to do. I literally just asked it to find a file, and it keeps loading forever in most of my projects 💀 They destroyed Grok 😭 https://t.co/3fLB0BwkzGview source ↗
-
• twitter 1w negativeCongrats on the round. Messaging is the distribution layer agents were missing — write the logic once, show up where people already talk. The npm numbers make that bet look real. Congrats on the round. Messaging is the distribution layer agents were missing — write the logic once, show up where people already talk. The npm numbers make that bet look real.view source ↗
-
• twitter 1w negativeOne senior AI engineer plus agents outperforms a five-person team. A PE-backed healthcare company had five people on a module rebuild. We put an AI Velocity Pod on comparable scope in the same codebase. Inside the pod: 1. One senior full-stack engineer paired with AI agents across the SDLC with an orbit of services 2… One senior AI engineer plus agents outperforms a five-person team. A PE-backed healthcare company had five people on a module rebuild. We put an AI Velocity Pod on comparable scope in the same codebase. Inside the pod: 1. One senior full-stack engineer paired with AI agents across the SDLC with an orbit of services 2. A half-time AI delivery architect covering review, architecture, and quality gates Total: 1.5 FTE at $15K-$20K a month. The traditional squad: two backend, two frontend, QA, PM, DevOps. Loaded cost between $610K and $880K a year. Week one goes entirely to building a knowledge graph of the codebase. Every module, dependency, data flow, and domain term stored where agents can reference it before generating anything. After that, every ticket follows six steps: define, spec, plan, implement, test, document. Agents generate roughly 98% of the code. The senior engineer reviews every PR through a gate we call V.U.E.: verify it works, understand why, explain it without the agent. Nothing merges into production unless the engineer passes all three. After 90 days running on the same scope: 1. Development cycles ran 50% faster 2. 122 merged PRs in 90 days 3. 85% of PRs needed fewer than 5 reviewer comments 4. Deployment cadence shifted from weeks to days 5. AI compute cost $200 per developer per month The traditional squad had one person writing code, three waiting on dependencies, and the fifth triaging blockers full time. One condition makes a Velocity Pod work or fail. The engineer has to be senior enough to catch what the agent gets wrong. Every buyer I talk to frames the same constraint: finding a senior developer who understands their domain takes longer than any other hire. AI raised the premium on seniority. AI Velocity Pod puts that person exactly where they compound: reviewing, directing, and deciding while agents handle volume.view source ↗
-
• twitter 1w negativeGiving an AI agent API access shouldn’t mean hand-wiring every integration. Airbyte Agent SDK is a type-safe connector execution framework for builders who want AI agents to work with external systems. It helps you connect agents to 50+ third-party APIs through strongly typed, documented tools that can run through th… Giving an AI agent API access shouldn’t mean hand-wiring every integration. Airbyte Agent SDK is a type-safe connector execution framework for builders who want AI agents to work with external systems. It helps you connect agents to 50+ third-party APIs through strongly typed, documented tools that can run through the Airbyte platform or locally in OSS mode. Key features: • Typed connector tools – gives agents documented interfaces and IDE autocomplete for connector operations • Prebuilt tool builder – turns a connector into inspect, docs, and execute tools for supported frameworks • Progressive documentation flow – guides an agent to inspect a connector and read relevant docs before execution • Framework-aware error handling – translates failures for PydanticAI, LangChain, OpenAI Agents, FastMCP, or a custom dispatch loop • Runtime safeguards – supports retries for transient failures and guards oversized tool output Public GitHub repo under the Elastic License 2.0. Link in the reply 👇view source ↗
-
• twitter 1w negativePurchased a Kling AI subscription for one month. one week left, and I couldn’t use even 50% of the credits. Most importantly, I couldn’t get good results from Instagram for app marketing. Unsure whether to renew it. https://t.co/xewARRksTy Purchased a Kling AI subscription for one month. one week left, and I couldn’t use even 50% of the credits. Most importantly, I couldn’t get good results from Instagram for app marketing. Unsure whether to renew it. https://t.co/xewARRksTyview source ↗
-
• twitter 1w negativeToday in AI Engineering (October 4) • Claude Code Mods: walkthrough details unsandboxed plugins with user permissions • Humanlayer releases Pi and OpenCode plugins for remote collaboration • LlamaIndex releases Extract v2.5, reports over 2× lower scanned-form errors • Google releases VeriHarness's 26,000 agent rollout… Today in AI Engineering (October 4) • Claude Code Mods: walkthrough details unsandboxed plugins with user permissions • Humanlayer releases Pi and OpenCode plugins for remote collaboration • LlamaIndex releases Extract v2.5, reports over 2× lower scanned-form errors • Google releases VeriHarness's 26,000 agent rollouts for shared-error checks • Hermes: Teknium details nearly 500 plugins, 8 officially tested, malware-only checks • Matt Pocock releases skills v1.3 with /retro reviews of old agent transcripts • Pi Durable: Raunak posts self-redeployment demo with SQLite memory and chat history • Pi Durable: Mario Zechner posts native Android prototype without Termux • Geoffrey Huntley posts self-modifying SBCL kernel demo with hot reload and rollback • NerfBench posts Opus 5.5 test methodology, cites ±10% as normal variance • Opus: Peter Gostev reports roughly 1500 to 1750 Elo in agent chess, with measurement errorview source ↗
-
• twitter 1w negativeI'm forced to use Astra in Codex since Sol 6.1 has some capacity issues today I'm forced to use Astra in Codex since Sol 6.1 has some capacity issues todayview source ↗
-
• twitter 1w negative@addyosmani Slow or unreliable test loops teach agents to just retry instead of fixing the root problem. Fast, deterministic checks keep the feedback cycle focused on real fixes. @addyosmani Slow or unreliable test loops teach agents to just retry instead of fixing the root problem. Fast, deterministic checks keep the feedback cycle focused on real fixes.view source ↗
-
• twitter 1w negativeTo make your agent produce high-quality output, equip it with a way to check its work. Tests are part of that. Here are some of the ones I think are worth investing in. - Write end-to-end tests that simulate real user flows and act as the ground truth - Use property-based testing to specify what should never happen a… To make your agent produce high-quality output, equip it with a way to check its work. Tests are part of that. Here are some of the ones I think are worth investing in. - Write end-to-end tests that simulate real user flows and act as the ground truth - Use property-based testing to specify what should never happen and generate thousands of test cases that attempt to break those constraints - If replacing a system, compare the old and new versions using randomly selected inputs - Require tests to run quickly and deterministically, which reinforces the agent to retry when the loop is slow or unreliable (the agent may learn to retry, not fix the problem) Example tests say what should happen; properties say what must not.view source ↗
-
• twitter 1w negativeWe paid Meta 4X what we paid for the ENTIRE month of September - in just the first 5 days of October. 🤯 For us, the new @WhatsApp pricing works out to nearly a 25X increase. The irony? If your brand simply broadcasts messages and customers don’t engage, the impact may not be huge. So spam impact isn’t much. But i… We paid Meta 4X what we paid for the ENTIRE month of September - in just the first 5 days of October. 🤯 For us, the new @WhatsApp pricing works out to nearly a 25X increase. The irony? If your brand simply broadcasts messages and customers don’t engage, the impact may not be huge. So spam impact isn’t much. But if you’ve actually built a channel where customers reply, interact and engage, the economics change dramatically. Brands using WhatsApp for real customer engagement will need to seriously rethink the channel. At these costs, it simply isn’t sustainable.view source ↗
-
• twitter 1w negativeim really confused, which model is actually better? - opus 5.5 - gpt-6.1 sol im really confused, which model is actually better? - opus 5.5 - gpt-6.1 solview source ↗
-
• twitter 1w negativeKey takeaways for marketers: 1. Lead with restoration, not novelty — answer what's back and whether public rooms return before pitching anything new 2. Build every post around a real captured moment — unexpected talent, a friend-group fail, a late-night reaction — and let the interface explain itself in the backgroun… Key takeaways for marketers: 1. Lead with restoration, not novelty — answer what's back and whether public rooms return before pitching anything new 2. Build every post around a real captured moment — unexpected talent, a friend-group fail, a late-night reaction — and let the interface explain itself in the background 3. Cast creators as groups, not solo endorsers. The product only makes sense when several people are visibly sharing one activity together 4. Separate the funnel: surprising moments for attention, shared reactions for proof, one clear action for conversion, recurring room series for retention 5. Treat the current TikTok account as a listening post, not a growth channel, until the product is confirmed active again.view source ↗
-
• twitter 1w negativeHow do people manage to work with Claude Code? I have recently been using it again a bit more (I get access to a quota subscription from work so it seems wasteful not to use it, now that the models are good again). I have also used the same models (Opus and Sonnet 5.5) through other harnesses (Pi, OpenCode, and Copilot… How do people manage to work with Claude Code? I have recently been using it again a bit more (I get access to a quota subscription from work so it seems wasteful not to use it, now that the models are good again). I have also used the same models (Opus and Sonnet 5.5) through other harnesses (Pi, OpenCode, and Copilot). In comparison, Claude Code is so bad, barely usable. Am I doing something wrong?view source ↗
-
• twitter 1w negativeTonight is the end of the @join_ef Bangalore → SF cohort Round 1! ⏱️ ₹3.6L equity-free on joining, 14 weeks in Bangalore, 3 months in San Francisco, up to $250k if you pass. No cofounder, idea, company, or US address needed. And free to apply. Yes, 2 more selection rounds after this. But the seats stay the same, wit… Tonight is the end of the @join_ef Bangalore → SF cohort Round 1! ⏱️ ₹3.6L equity-free on joining, 14 weeks in Bangalore, 3 months in San Francisco, up to $250k if you pass. No cofounder, idea, company, or US address needed. And free to apply. Yes, 2 more selection rounds after this. But the seats stay the same, with more applicants and worse odds each time! Round 1 is the easiest shot you will ever get at this. Stop reading and go submit:view source ↗
-
• twitter 1w negativeClaude's auto mode classifier has gotten really fucking bad the number of legit and explicitly approved actions that the classifier denies regardless is honestly fucking ridiculous it makes the app borderline unusable it wasn't like that a few weeks ago what changed? @bcherny @amorriscode Claude's auto mode classifier has gotten really fucking bad the number of legit and explicitly approved actions that the classifier denies regardless is honestly fucking ridiculous it makes the app borderline unusable it wasn't like that a few weeks ago what changed? @bcherny @amorriscodeview source ↗
-
• twitter 1w negativeI had a crush on this boy who was a startup founder. And then he ghosted me and it made me really sad. So I started a startup and got into YC. 👸 I had a crush on this boy who was a startup founder. And then he ghosted me and it made me really sad. So I started a startup and got into YC. 👸view source ↗
-
• twitter 1w negative"AI content doesn't rank" is the wrong lesson. Google is using AI to write answers at the top of its own results. What Google ignores isn't AI. It's content that adds nothing new to the internet. Here's what actually ranks, based on Julian's setup: ✔ Every fact pulled from a real case study ✔ Your actual results a… "AI content doesn't rank" is the wrong lesson. Google is using AI to write answers at the top of its own results. What Google ignores isn't AI. It's content that adds nothing new to the internet. Here's what actually ranks, based on Julian's setup: ✔ Every fact pulled from a real case study ✔ Your actual results and numbers, nothing invented ✔ Clear structure, every sentence on its own line ✔ Smart internal and external links SEOs call this information gain. Google rewards it. 🧠 The numbers he reports: → One site: zero to 325 clicks a day → Another: up to 125 clicks a day → Timeline: 3 to 6 months for real results And one site stayed flat from April to November before it broke out of the sandbox. Most people quit at page 10. The ones who keep publishing are the ones who see the curve. Save this video, you'll need it the next time someone says AI content can't rank. Want the SOP? DM me.view source ↗
-
• twitter 1w negativeAccording to studies, high-anxiety people are more likely to make bad decisions because they tend to catastrophize uncertain situations. According to studies, high-anxiety people are more likely to make bad decisions because they tend to catastrophize uncertain situations.view source ↗
-
• twitter 1w negativeif you ever feel useless, remeber ai[dot]com burned $70M for literally nothing 😭 https://t.co/Csk3qTZrEG if you ever feel useless, remeber ai[dot]com burned $70M for literally nothing 😭 https://t.co/Csk3qTZrEGview source ↗
-
• twitter 1w negativetibo realizing he can't win the pr battle against lauren https://t.co/HFq7gd6XvF tibo realizing he can't win the pr battle against lauren https://t.co/HFq7gd6XvFview source ↗
-
• twitter 1w negativeCan't access GPT-6.1 Sol in Codex now 😭 What is happening https://t.co/SjwPHaixLn Can't access GPT-6.1 Sol in Codex now 😭 What is happening https://t.co/SjwPHaixLnview source ↗