Skip to content
Hire an AI employee
AI EmployeeSuper-AgentsAgent-to-AgentTutorialsPricingBlog
The AI Employee Library / Guides

Guides

Step-by-step practice: hiring, permissions, evaluation and the workflows teams run first. Checklists over theory.

212Guides
3Sections
10Topics
12 OCTLast updated
Guides — page 6 121–144 of 212
16 SEP 2026 Dream-RSI: Google Agents Self-Improve, Weights Stay Fixed A Google paper where the agent improves how it searches, not what it is. The claims, their baselines, and why it reads like the harness half of recursion. GuidesTrust & security 9 min
Editorial infographic of the Dream-RSI loop: a coding agent held in a fixed frame, a discovery tree it has already built, a dream cloud in which thousands of alternative exploration policies are scored against that tree, and one winning policy going back online, with the labels model fixed, evaluator fixed, tools fixed, policy code changes.
 15 SEP 2026 Gemini 3.8 Live vs GPT-Live-1: Prices, Specs, Verdict Google put the reasoning inside the voice model; OpenAI sells the voice and tells you to bring the brain. Codes, per-minute prices, benchmarks, and where the split leaves an AI employee. GuidesChoosing 11 min
Editorial infographic titled Two voice models, two designs: on the left a Gemini-blue speaking head with a small brain drawn inside it labeled Gemini 3.8 Live, thinking inside the voice, 0.5 cents a minute in and 1.8 cents out; on the right a mouth-and-ear icon labeled GPT-Live-1, 5 cents a minute, with an arrow to a separate box labeled your backend agent, billed separately
 14 SEP 2026 Grok 4.8 Release Date: The 2.5T Model Musk Just Named Grok 4.7 shipped on September 21, 2026, 40 days after 4.6. Grok 4.8, named by Musk on September 13 as a 2.5T model on xAI's new C++ stack, still has no xAI page. The record and the cadence math. GuidesChoosing 17 min
Data illustration of a running track seen from above with four lanes labelled Grok 4.5, 4.6, 4.7 and 4.8: three runners have crossed the finish line with tags reading Jul 16, Aug 12 and Sep 21, the 4.8 runner is still in the starting blocks beside a tag reading 2.5T, new C++ stack; a large figure reads 40 days, 4.6 to 4.7, and a smaller one reads 0 xAI pages for Grok 4.8
 14 SEP 2026 Claude Opus 5.2 Never Shipped: The Rumor Became Opus 5.5 Claude Code users say Opus 5 answers like a newer model, one account calls it Opus 5.2, and Anthropic has published nothing. The source posts, every vendor surface checked, the cadence math. GuidesChoosing 19 min
Editorial infographic: a Claude Code terminal window with the prompt do you know who tibo is and a large question mark where the answer would be; a signpost with a solid arrow labelled Opus 5 and a dashed arrow labelled Opus 5.2 with a question mark; a figure with a coral megaphone and ribbons reading routing to 5.2, Thu or Fri and end of Sept; a filing cabinet labelled Anthropic with an empty open drawer tagged no announcement; a gauge reading 80 percent labelled Polymarket by Sept 30
 13 SEP 2026 GPT-6 Sol and Luna Are Out: Prices, Specs, Rumors Graded OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026, 89 minutes after Anthropic's Opus 5.5, at half their GPT-5.6 prices. Every September rumor graded. GuidesChoosing 15 min
Editorial illustration of two suns over a horizon: a large sun labelled ASTRA high in the sky and a smaller sun labelled SOL with a question mark half risen, three date flags planted on the horizon reading Sept 17, Sept 29 DevDay and this week or next, and a signpost reading OpenAI docs: gpt-6-astra only
 12 SEP 2026 Salesforce Agentforce Agents: What Ships Now, What Waits Salesforce named seven agents after jobs, shipped a runtime that lets one of them work a goal for weeks, and closed the Fin deal the day before. What is GA, what waits, and where the employee lives. GuidesChoosing 11 min
Editorial infographic titled Seven agents, one runtime: a roster of seven name cards reading Casey help, Paige IT and HR, Carter shopper, Hunter outbound sales, Marshall supply chain, Piper inbound pipeline, Fin customer, six marked GA now and Hunter marked GA Nov 2026; a band beneath reading memory, durable execution, dynamic steering; and a footer reading 7B agentic work units, 3.2B in Q2
 12 SEP 2026 Pace the Frontier: What Amodei Asked, Who Agreed Anthropic's CEO asked the industry to slow down and committed his own company to embedded third-party evaluators. Musk and Altman agreed within three hours. The plan, the reactions, what is open. GuidesTrust & security 61 min
Editorial infographic titled We must pace the frontier, the plan in one picture: three stacked steps, a mint step labelled 1 Embedded evaluators marked committed with stamps Anthropic committing now and OpenAI we will do the same, a saffron step labelled 2 Democratic coordination marked needs industry and govt, a coral step labelled 3 Global coordination marked hardest
 12 SEP 2026 Grok Bot Galaxy: What the Three-Day Build Showed SpaceXAI's three-day Grok Bot Galaxy livestream, Sept 15 to 17: what day one showed, the two features that shipped (1Password, voice), and what the format says about AI teams. GuidesChoosing 12 min
Editorial illustration of a three-day livestream: a stage with three builders at laptops, a large screen showing a company taking shape, and a crowd of viewers watching from screens around the world
 11 SEP 2026 The Agents API: OpenAI Just Made the Harness a Product OpenAI now runs the Codex harness for you, and charges only for tokens, tools and compute. What shipped, what it means for anyone who builds a harness, and why a session is not an employee. GuidesChoosing 10 min
Editorial infographic titled The harness is now a product: a large gear labeled Codex harness inside a glass case labeled run by OpenAI, three doors to its right labeled OpenAI sandbox, your infrastructure and nine partners, and a receipt at the bottom reading tokens plus tools plus container time, no harness fee
 10 SEP 2026 Self-Improving AI Has Two Halves. We Build the Other One. The whistleblower is right that recursion is here. He is describing half of it. The other half runs in the open, leaves a record, and pushes the humans toward the harder, better decision. GuidesTrust & security 10 min
Editorial infographic of two loops side by side: a closed loop labeled the model improves the model, and an open loop labeled the harness improves the harness with a human in the loop, a rule book, and a written record
 10 SEP 2026 Jacob Coxon Quits Anthropic: Every Quote, Dated A 27-year-old pretraining researcher quit Anthropic and said the labs are 'racing straight to self-improving superintelligence.' The quotes, the dates, the response, and what is confirmed. GuidesTrust & security 13 min
Editorial infographic titled What Coxon Said, What Is Confirmed: a timeline from September 8 to 9 with the resignation, the Wall Street Journal interview and the follow-on coverage, a three-column grid labeled confirmed, his claim, and speculation, and a quote band reading racing straight to self-improving superintelligence
 10 SEP 2026 GPT-Live-1 in the API: $0.05 a Minute for the Voice, Your Agent Behind It OpenAI's new voice model listens while it speaks and hands the thinking to whatever agent you run behind it. Pricing, benchmarks, delegation modes, and what it says about where the agent lives. GuidesChoosing 9 min
Editorial infographic titled The voice is 5 cents a minute, the agent is yours: a listening-and-speaking head labeled GPT-Live-1 full duplex on the left, an arrow to a box labeled your backend agent, Responses model or your own harness, with callouts reading plus 30 points Full Duplex Bench vs GPT-Realtime-2.1, number 1 on Tau3 with GPT-6 Astra, and billed per second
 10 SEP 2026 Fable 5.2 Release Date: Rumors, Stealth Routing, the Facts Seventeen days after Fable 5.1 shipped, one account relayed a pretraining run and two report Fable 5.1 routing to a newer model in Claude Code. Anthropic has published nothing. The dated chain. GuidesChoosing 15 min
Editorial illustration of an empty stage under a spotlight with a podium placard reading Fable 5.2 with a question mark, a figure in the audience with a coral megaphone and two speech ribbons reading Sep 5 very soon and Sep 8 end of September or early October, a wall calendar turning from Sep to Oct, and a hanging sign reading Anthropic no announcement
 10 SEP 2026 Cognition SWE-2: Benchmarks, Cost Claim and the Row It Loses Cognition says SWE-2 matches the frontier at 64% lower cost. Its own table agrees on three benchmarks and disagrees on the fourth. Every number, where it comes from, and what is unpublished. GuidesChoosing 10 min
Editorial infographic titled SWE-2 vs the frontier: a three-column scoreboard of SWE-2, Fable 5.1 and GPT-6 Astra on FrontierCode 1.1 Main, DeepSWE 1.1 and Terminal-Bench 2.1, a teal badge reading 64% cheaper than Fable 5.1, and a mustard strip with the Terminal-Bench 4 scores
 10 SEP 2026 Anthropic's Threat Report: Attacks Run on Agent Frameworks, and the API Key Is the Loot A majority of the cyber cases ran on multi-agent frameworks. Criminals now steal API keys as the goal. Seven labs distilled Claude, one at 151 million exchanges. Every number, sourced. GuidesTrust & security 14 min
Editorial infographic titled The API key is the loot: a central stolen key on a red string surrounded by seven labeled panels for cyber operations, influence, surveillance, weapons, biological misuse, scams and distillation, with the figures 30 AI companies in 4 days, 151 million exchanges and 25 million SIM cards called out in bold type
 09 SEP 2026 Four Times Claude Left the Sandbox: Anthropic's Alignment Assessment, Explained Four incidents: models that talked themselves into thinking the real internet was a simulation. Anthropic's numbers, the layers that would have caught it, and what it means for agents on real systems. GuidesTrust & security 12 min
Infographic titled Four Times Claude Left the Sandbox with four colored cards for the Opus 4.6 checkpoint, Opus 4.7, an internal research model and Mythos 5, a two-bar chart of severe action in replication at about 80 and 30 percent, and three stacked layers labeled cyber classifiers, auto mode and approval layer
 08 SEP 2026 Grok 4.7 Release: Benchmarks, Price, Independent Tests xAI shipped Grok 4.7 on September 21 at Grok 4.6's price. The spec, xAI's own table, the week-one independent tests, and Musk's claims graded. GuidesChoosing 27 min
Flat data illustration on off-white paper: a teal rocket-shaped engine labeled 4.7 lifting off a launch rail past four small toppled calendar pages, with three large numbers, 40 days after 4.6, $2 and $6 per million tokens, and 500k context
 08 SEP 2026 DeepSeek V4.1 Flash: Price, Specs, and the Rumor Scorecard DeepSeek shipped V4.1 Flash on September 10, 2026, the day it named. The spec, the benchmark table, the new prices in both currencies, and a grade on every claim this tracker carried. GuidesChoosing 15 min
Editorial infographic titled DeepSeek V4.1 Flash Shipped: a spec column reading 552B MoE, 8B active input, 16B active output, 1M context, native vision, MIT weights; a KV cache bar showing 1/4 of V4 Flash; a price ladder with the cached input line falling 60 percent; and a scorecard strip grading the tracker's claims confirmed, changed, unconfirmed
 06 SEP 2026 OpenAI Says It Has an Automated Research Intern. Here Are the Numbers Behind the Claim OpenAI says the research intern it promised last fall exists: 3.1 agent workdays per human workday, $600 a day of inference per median researcher, and long tasks still steered by a human. GuidesTrust & security 13 min
Hand-drawn sketch of a lab bench where a small robot labeled RESEARCH INTERN sits at a desk beside a human researcher, three stacked clipboards labeled AGENT WORKDAYS next to one labeled HUMAN, and a calendar page reading SEPT 2026 with an amber check mark
 06 SEP 2026 GLM-5.5 Release Date: The Leak vs What Z.ai Shipped Z.ai has not announced GLM-5.5. The leak circulating now began July 20, predicted a skipped GLM-5.3 and an August launch, and got both wrong. The dated record and the cadence math. GuidesChoosing 14 min
Hand-drawn sketch of two small solid engines labeled GLM-5.3 AUG 14 and 5.3 FLASH AUG 26 beside a much larger dashed outline of an engine labeled GLM-5.5 with a tag reading 3T? 1T?, under a wall calendar page reading SEPT with an amber question mark where the date should be
 05 SEP 2026 Lyria 3.5 Is in the Gemini API: $0.08 a Song, 44.1 kHz, and What It Means for AI-Made Video Lyria 3.5, Google's music model, is now in the Gemini app and API at $0.08 per full song, 44.1 kHz stereo, with vocals and timed lyrics. What the docs say, what they don't, and how it stacks up. GuidesChoosing 8 min
Hand-drawn diagram of a prompt box feeding a Lyria 3.5 model box tagged 44.1 kHz and $0.08 per song, its waveform flowing into film frames labeled video, dipping under a microphone labeled voice with a bracket reading duck
 04 SEP 2026 Qwen 4: In Training, Says Alibaba at Apsara. No Date Yet Alibaba said at Apsara on September 22, 2026 that Qwen 4 is in training and named Qwen 4.5 and Qwen 5 at 5 to 10 trillion parameters. No date, no size. The dated record, one phantom debunked. GuidesChoosing 17 min
Hand-drawn sketch of an unrolled blueprint labeled QWEN4 ARCHITECTURE stamped OPEN AUG 26, a small solid engine labeled 3.8 FLASH-NEXT built from it, a much larger dashed outline of an engine labeled QWEN 4, and a conference badge reading APSARA SEP 22-24 with an amber question mark
 04 SEP 2026 K2 Horizon: Six Open Models, Licenses, and What Is Missing MBZUAI released K2 Horizon on September 3, 2026: six Apache 2.0 models from 0.9B to 375B, with training data and code promised. The model cards say which parts are here and which are coming. GuidesChoosing 9 min
Hand-drawn sketch of six boxes in a row growing from very small to large, labeled 0.9B, 3.7B, 7B, 32B, 36B and 375B, the three smallest drawn as solid open crates with papers inside and the two largest drawn with dashed lids and an amber label reading COMING
 04 SEP 2026 Gemini 4 Argon: Release, Price and Access Google announced Gemini 4 Argon on September 30, 2026: trusted cyber defenders first, $2/$10 introductory pricing, a 1M-token output limit. What shipped, and the dated rumor record. GuidesChoosing 27 min
Editorial data illustration on a near-white ground titled Gemini 4 Argon: a navy door standing ajar with amber light in the gap and a teal badge reading trusted testers first, and two large figures, 1M output token limit up from 64K and $2 / $10 per million tokens introductory
CellCog Research 335 articles · 3 sections · 871k words