# Jev Daily

Jev, read daily so you do not have to. The builds, the benchmarks, the criteria that worked and the cases where it lost, from the people shipping on TypeSafe AI's System One model.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## PostHog puts 444.6x cheaper on decision models

DevFeed: [PostHog puts 444.6x cheaper on decision models](<https://devfeed.tech/articles/posthog-puts-444-6x-cheaper-on-decision-models-146091.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-20>)

Author: Jev Daily

Published: 2026-11-20T14:11:27.545612Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [antigravity](<https://devfeed.tech/tags/antigravity.md>), [decision-models](<https://devfeed.tech/tags/decision-models.md>), [faster](<https://devfeed.tech/tags/faster.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [gemini-4-neon](<https://devfeed.tech/tags/gemini-4-neon.md>), [gpt-6-2-astra](<https://devfeed.tech/tags/gpt-6-2-astra.md>), [jev](<https://devfeed.tech/tags/jev.md>), [kitchen](<https://devfeed.tech/tags/kitchen.md>), [llm](<https://devfeed.tech/tags/llm.md>), [opus-5-5](<https://devfeed.tech/tags/opus-5-5.md>), [prompt-caching](<https://devfeed.tech/tags/prompt-caching.md>), [run](<https://devfeed.tech/tags/run.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [temperature](<https://devfeed.tech/tags/temperature.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [theo](<https://devfeed.tech/tags/theo.md>), [waiting](<https://devfeed.tech/tags/waiting.md>), [work](<https://devfeed.tech/tags/work.md>)

### AI overview

PostHog reports that Jev, a decision model, handles some work up to 444.6 times cheaper and 193.6 times faster than constant LLM calls. The excerpt does not identify the workload, comparison LLM, or measurement method, so the figures cannot be assessed further from this evidence. Separately, Theo uses a waiter analogy to criticize an agent that pauses a task to ask an unnecessary question; he provides no run, model settings, or task, so this is not a measured result.

### Source excerpt

PostHog says decision models like Jev do some of its work up to 444.6x cheaper and 193.6x faster than constant LLM calls, without saying which work or against...

## Rivet split one Jev question into six

DevFeed: [Rivet split one Jev question into six](<https://devfeed.tech/articles/rivet-split-one-jev-question-into-six-146090.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-19>)

Author: Jev Daily

Published: 2026-11-19T17:00:00Z

Content type: article

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [if statement](<https://devfeed.tech/topics/if-statement.md>)

Tags: [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [classifier](<https://devfeed.tech/tags/classifier.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [engineers](<https://devfeed.tech/tags/engineers.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [jev](<https://devfeed.tech/tags/jev.md>), [question](<https://devfeed.tech/tags/question.md>), [results](<https://devfeed.tech/tags/results.md>), [split](<https://devfeed.tech/tags/split.md>), [support](<https://devfeed.tech/tags/support.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [ticket](<https://devfeed.tech/tags/ticket.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Rivet found that Jev classified ticket severity poorly when asked to choose SEV1-SEV4 in one step. Asking three yes/no questions--whether the issue is down, blocked, or broken--and ranking the answers with simple if-statements produced a working classifier. Nathan Flurry says Rivet built its support system with Rivet and Jev, reducing time to resolve hard engineering tickets to 5h 1m; human engineers still staff support, with models assisting. The excerpt also notes that Rivet had previously used Plain, which Flurry called its best attempt so far.

### Source excerpt

Nathan Flurry says Rivet asked Jev for a ticket's severity in one question and got mediocre results, then got a working classifier by asking three yes/no...

## Treg's open jobs API runs a job hunt for $0.03

DevFeed: [Treg's open jobs API runs a job hunt for $0.03](<https://devfeed.tech/articles/treg-s-open-jobs-api-runs-a-job-hunt-for-0-03-142989.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-18>)

Author: Jev Daily

Published: 2026-11-18T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [API Business Models](<https://devfeed.tech/topics/api-business-models.md>), [llamaindex](<https://devfeed.tech/topics/llamaindex.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [api](<https://devfeed.tech/tags/api.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [call](<https://devfeed.tech/tags/call.md>), [costs](<https://devfeed.tech/tags/costs.md>), [data](<https://devfeed.tech/tags/data.md>), [decision](<https://devfeed.tech/tags/decision.md>), [hiring](<https://devfeed.tech/tags/hiring.md>), [ignite](<https://devfeed.tech/tags/ignite.md>), [input](<https://devfeed.tech/tags/input.md>), [jev](<https://devfeed.tech/tags/jev.md>), [job](<https://devfeed.tech/tags/job.md>), [job-hunt](<https://devfeed.tech/tags/job-hunt.md>), [jobs](<https://devfeed.tech/tags/jobs.md>), [latency](<https://devfeed.tech/tags/latency.md>), [markup](<https://devfeed.tech/tags/markup.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [open](<https://devfeed.tech/tags/open.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [published](<https://devfeed.tech/tags/published.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>)

### AI overview

Treg's open-source Jev-powered jobs API gathers job data through one endpoint for use by an agent. He says a full job hunt costs $0.03 and returns salary and hiring-manager data; it is published with 0% markup. Separately, OpenRouter says Microsoft-Decision-1.1 is generally available: it returns calibrated probabilities for answer options in one call, in the same output shape Jev uses. The excerpt gives no accuracy, latency, or price figures for that model.

### Source excerpt

Treg says a full job hunt through his new Jev-powered jobs API costs three cents, open source at 0% markup. Meanwhile Microsoft put a calibrated decision...

## Browser Use blocked 100/100, and batch your Jev calls

DevFeed: [Browser Use blocked 100/100, and batch your Jev calls](<https://devfeed.tech/articles/browser-use-blocked-100-100-and-batch-your-jev-calls-137103.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-16-w7>)

Author: Jev Weekly

Published: 2026-11-16T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Browser Automation](<https://devfeed.tech/topics/browser-automation.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [also-on-the-timeline](<https://devfeed.tech/tags/also-on-the-timeline.md>), [api](<https://devfeed.tech/tags/api.md>), [autoresearch](<https://devfeed.tech/tags/autoresearch.md>), [batching](<https://devfeed.tech/tags/batching.md>), [browser](<https://devfeed.tech/tags/browser.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [bun](<https://devfeed.tech/tags/bun.md>), [captcha](<https://devfeed.tech/tags/captcha.md>), [captchas](<https://devfeed.tech/tags/captchas.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [context-window](<https://devfeed.tech/tags/context-window.md>), [decisions-api](<https://devfeed.tech/tags/decisions-api.md>), [dependency](<https://devfeed.tech/tags/dependency.md>), [detection](<https://devfeed.tech/tags/detection.md>), [editor](<https://devfeed.tech/tags/editor.md>), [fable-5](<https://devfeed.tech/tags/fable-5.md>), [gemini-decisions-api](<https://devfeed.tech/tags/gemini-decisions-api.md>), [hetzner](<https://devfeed.tech/tags/hetzner.md>), [ilias](<https://devfeed.tech/tags/ilias.md>), [jev](<https://devfeed.tech/tags/jev.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

A report says Browser Use was blocked in 100 of 100 separate runs, while another test blocked all 5 fresh Browser Use Cloud sessions on US residential proxies using solveCaptchas=true; rotate-image captchas remained unsolved after 120 seconds. Separately, Jev's input pricing charges for state once plus the questions, making parallel batching cheaper than sequential calls that resend state. An account of a self-hosted Postiz alternative describes a single Bun process with bun:sqlite WAL, 20 MCP tools across 27 providers, and $0 monthly cost on an existing Hetzner server. Other excerpted reports cover Gemini Decisions API pricing and benchmark claims, and a context-summary skill tuned to match its author's taste 80% of the time.

### Source excerpt

sahir says one bot gate blocked Browser Use on 100 of 100 separate runs, and Gregor Zunic agreed it got them 5/5 on his own attempt. The cheaper lesson of the...

## Riley Brown's 4K film: 89 rented GPUs, $20

DevFeed: [Riley Brown's 4K film: 89 rented GPUs, $20](<https://devfeed.tech/articles/riley-brown-s-4k-film-89-rented-gpus-20-135260.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-15>)

Author: Jev Daily

Published: 2026-11-15T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [blender](<https://devfeed.tech/topics/blender.md>), [3D](<https://devfeed.tech/topics/3d.md>), [Python](<https://devfeed.tech/topics/python.md>), [prompt](<https://devfeed.tech/topics/prompt.md>), [GPUDirect](<https://devfeed.tech/topics/gpudirect.md>)

Tags: [4k](<https://devfeed.tech/tags/4k.md>), [agent](<https://devfeed.tech/tags/agent.md>), [blender](<https://devfeed.tech/tags/blender.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [fable-5-2](<https://devfeed.tech/tags/fable-5-2.md>), [film](<https://devfeed.tech/tags/film.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [parallel](<https://devfeed.tech/tags/parallel.md>), [project](<https://devfeed.tech/tags/project.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [python](<https://devfeed.tech/tags/python.md>), [questions](<https://devfeed.tech/tags/questions.md>), [render](<https://devfeed.tech/tags/render.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [state](<https://devfeed.tech/tags/state.md>), [weekend-project](<https://devfeed.tech/tags/weekend-project.md>)

### AI overview

Riley Brown says a single Fable 5.2 prompt generated 12,500 lines of Blender Python for a 2-minute 4K film with 5 characters, 7 sets and 52 shots. Rendering 3,220 frames used 89 rented GPUs and cost a little over $20.

### Source excerpt

Riley Brown says one Fable 5.2 prompt wrote 12,500 lines of Blender Python, then rented 89 GPUs to render 3,220 frames for a little over $20. Also: Allie the...

## Batch the Jev questions you might not need

DevFeed: [Batch the Jev questions you might not need](<https://devfeed.tech/articles/batch-the-jev-questions-you-might-not-need-134347.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-14>)

Author: Jev Daily

Published: 2026-11-14T17:00:00Z

Content type: opinion

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>)

Tags: [batch](<https://devfeed.tech/tags/batch.md>), [batching](<https://devfeed.tech/tags/batching.md>), [browser](<https://devfeed.tech/tags/browser.md>), [cache](<https://devfeed.tech/tags/cache.md>), [call](<https://devfeed.tech/tags/call.md>), [cost](<https://devfeed.tech/tags/cost.md>), [eval](<https://devfeed.tech/tags/eval.md>), [faster](<https://devfeed.tech/tags/faster.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [harness](<https://devfeed.tech/tags/harness.md>), [hit](<https://devfeed.tech/tags/hit.md>), [jev](<https://devfeed.tech/tags/jev.md>), [know](<https://devfeed.tech/tags/know.md>), [logic](<https://devfeed.tech/tags/logic.md>), [minutes](<https://devfeed.tech/tags/minutes.md>), [numbers](<https://devfeed.tech/tags/numbers.md>), [one](<https://devfeed.tech/tags/one.md>), [questions](<https://devfeed.tech/tags/questions.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [state](<https://devfeed.tech/tags/state.md>), [tokens](<https://devfeed.tech/tags/tokens.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Allie the Icon recommends batching Jev questions into one call even when you may not use every answer. The stated cost rationale is that state tokens are paid once per call, whereas sequential calls on different logic paths duplicate that state. Separately, Browser Use reports that fixing its eval harness raised cache hit rate from 78% to 96%; cost fell from $0.67 to $0.45 per task and time from 5.0 to 4.6 minutes, with the same score. The fix shipped in browser-use 1.0.4. The excerpt gives no further implementation details for either practice.

### Source excerpt

Allie the Icon says batching Jev questions you may never use is usually faster and cheaper, because state tokens are paid once per call and questions do not...

## Jev updates a voice order ~200ms after you speak

DevFeed: [Jev updates a voice order ~200ms after you speak](<https://devfeed.tech/articles/jev-updates-a-voice-order-200ms-after-you-speak-130278.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-11>)

Author: Jev Daily

Published: 2026-11-11T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Conversational AI](<https://devfeed.tech/topics/conversational-ai.md>), [Chat Bot](<https://devfeed.tech/topics/chatbot.md>)

Tags: [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [bill](<https://devfeed.tech/tags/bill.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [coffee](<https://devfeed.tech/tags/coffee.md>), [frontier-model](<https://devfeed.tech/tags/frontier-model.md>), [jev](<https://devfeed.tech/tags/jev.md>), [order](<https://devfeed.tech/tags/order.md>), [osint](<https://devfeed.tech/tags/osint.md>), [screen](<https://devfeed.tech/tags/screen.md>), [sextant](<https://devfeed.tech/tags/sextant.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [tokens](<https://devfeed.tech/tags/tokens.md>), [voice](<https://devfeed.tech/tags/voice.md>)

### AI overview

Kunal Abichandani says his Jev-powered coffee ordering system renders a drink and requested changes about 200 ms after the words are spoken, while the customer is still talking; he does not describe how latency was measured or the available choices. Everton Carneiro reports spending $8.04 for 201 million tokens using Jev for keyword judging in Sextant, arguing that not every call in a stack needs a frontier model. Separately, ILIAS ISM says he used an AI agent to run OSINT and submit abuse reports after a customer sought a refund on a product clone.

### Source excerpt

Kunal Abichandani says his voice coffee ordering system puts the drink on screen about 200ms after the word, while Everton Carneiro says 201 million tokens of...

## Mike Taylor's 65 builds and Theo's $42 Fable run

DevFeed: [Mike Taylor's 65 builds and Theo's $42 Fable run](<https://devfeed.tech/articles/mike-taylor-s-65-builds-and-theo-s-42-fable-run-128949.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-10>)

Author: Jev Daily

Published: 2026-11-10T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Fable](<https://devfeed.tech/topics/fable.md>), [3D](<https://devfeed.tech/topics/3d.md>)

Tags: [3d](<https://devfeed.tech/tags/3d.md>), [agent](<https://devfeed.tech/tags/agent.md>), [better](<https://devfeed.tech/tags/better.md>), [browser](<https://devfeed.tech/tags/browser.md>), [builds](<https://devfeed.tech/tags/builds.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [cdp](<https://devfeed.tech/tags/cdp.md>), [chrome](<https://devfeed.tech/tags/chrome.md>), [chrome-devtools](<https://devfeed.tech/tags/chrome-devtools.md>), [fable](<https://devfeed.tech/tags/fable.md>), [fable-5-2](<https://devfeed.tech/tags/fable-5-2.md>), [game](<https://devfeed.tech/tags/game.md>), [minutes](<https://devfeed.tech/tags/minutes.md>), [models](<https://devfeed.tech/tags/models.md>), [opus](<https://devfeed.tech/tags/opus.md>), [opus-5-5](<https://devfeed.tech/tags/opus-5-5.md>), [performance](<https://devfeed.tech/tags/performance.md>), [python](<https://devfeed.tech/tags/python.md>), [run](<https://devfeed.tech/tags/run.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [theo](<https://devfeed.tech/tags/theo.md>), [tokens](<https://devfeed.tech/tags/tokens.md>)

### AI overview

Browser Use 1.0.0 replaces its predefined state and action space with an agent-written Python REPL operating against raw CDP. The team reports that, using the same models, this approach was up to 50% better, 2-5x cheaper, and 2-3x faster. Mike Taylor logged 65 builds while spending about 14 billion tokens testing Fable 5.2 over four weeks. Theo says Fable built his game in 76 minutes for about $42 in tokens, but found it not substantially better than Opus 5.5 for the price; he rated the Opus version as more playable, with better controls and performance. He sees the models as potentially complementary for 3D game work.

### Source excerpt

Mike Taylor spent about 14 billion tokens over four weeks and found Jev labelling an hour of Twitch video for 14 cents against roughly $6 for his 2024...

## TypeSafe: Jev is not a decision model

DevFeed: [TypeSafe: Jev is not a decision model](<https://devfeed.tech/articles/typesafe-jev-is-not-a-decision-model-125629.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-09-w7>)

Author: Jev Weekly

Published: 2026-11-09T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [AI Integration Strategies](<https://devfeed.tech/topics/ai-integration-strategies.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Prompt Engineering](<https://devfeed.tech/topics/prompt-engineering.md>), [decision-making](<https://devfeed.tech/topics/decision-making.md>)

Tags: [agent-skills](<https://devfeed.tech/tags/agent-skills.md>), [also-on-the-timeline](<https://devfeed.tech/tags/also-on-the-timeline.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [code](<https://devfeed.tech/tags/code.md>), [decision-model](<https://devfeed.tech/tags/decision-model.md>), [jev](<https://devfeed.tech/tags/jev.md>), [reliability](<https://devfeed.tech/tags/reliability.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [typesafe](<https://devfeed.tech/tags/typesafe.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

The excerpt argues that reliability improves when code makes final decisions and Jev handles judgments or other tasks. It describes an MCP spending safeguard: ILIAS ISM's backlink tool previews the quote and charges only after a separate call with `confirmPaymentSpend: true`. Milind S reports a bug-fixing workflow in which a cloud agent tests changes on a virtual machine and attaches a screenshot to a pull request for team review. Moritz Kremb's skills read a feedback document on each run, allowing updates without changing the skill itself. Mike Taylor reports decision-clone accuracy of 60% after one shot, 85% after ten, and 95% after 400. Max Blade proposes automating recurring tasks in the agent harness, but the excerpt provides no implementation or performance numbers.

### Source excerpt

TypeSafe AI spent the week arguing its own model should never be called a decision model, because reliability comes from code composing judgments. Meanwhile...

## Mike Taylor: 400 shots to clone a decision at 95%

DevFeed: [Mike Taylor: 400 shots to clone a decision at 95%](<https://devfeed.tech/articles/mike-taylor-400-shots-to-clone-a-decision-at-95-122447.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-07>)

Author: Jev Daily

Published: 2026-11-07T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>), [decision-making](<https://devfeed.tech/topics/decision-making.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [api](<https://devfeed.tech/tags/api.md>), [clone](<https://devfeed.tech/tags/clone.md>), [context](<https://devfeed.tech/tags/context.md>), [data](<https://devfeed.tech/tags/data.md>), [decision](<https://devfeed.tech/tags/decision.md>), [faster](<https://devfeed.tech/tags/faster.md>), [model](<https://devfeed.tech/tags/model.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [repeat](<https://devfeed.tech/tags/repeat.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [team](<https://devfeed.tech/tags/team.md>), [testing](<https://devfeed.tech/tags/testing.md>), [writing](<https://devfeed.tech/tags/writing.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Mike Taylor reports that cloning Dan's decisions reached 60% accuracy with one example, 85% with ten, and 95% with 400. Alexander proposed testing on held-out real decisions to check whether clones reproduce the team's reasoning rather than its writing style; Taylor called this "phase 2." Separately, Greg Isenberg describes a company loop of acquiring valuable context, orchestrating agents, feeding data back, and involving humans, claiming that context and loop speed matter more than model quality.

### Source excerpt

Mike Taylor says cloning one person's decisions lands at 60% accuracy one shot and needs 400 shots to reach 95%. If you are planning to clone your team's...

## TypeSafe: Jev is not a decision model

DevFeed: [TypeSafe: Jev is not a decision model](<https://devfeed.tech/articles/typesafe-jev-is-not-a-decision-model-121990.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-06>)

Author: Jev Daily

Published: 2026-11-06T17:00:00Z

Content type: opinion

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [AI Integration Strategies](<https://devfeed.tech/topics/ai-integration-strategies.md>)

Tags: [abstraction](<https://devfeed.tech/tags/abstraction.md>), [agent](<https://devfeed.tech/tags/agent.md>), [all](<https://devfeed.tech/tags/all.md>), [button](<https://devfeed.tech/tags/button.md>), [code](<https://devfeed.tech/tags/code.md>), [decision-model](<https://devfeed.tech/tags/decision-model.md>), [harness](<https://devfeed.tech/tags/harness.md>), [implementation](<https://devfeed.tech/tags/implementation.md>), [jev](<https://devfeed.tech/tags/jev.md>), [layer](<https://devfeed.tech/tags/layer.md>), [max](<https://devfeed.tech/tags/max.md>), [model](<https://devfeed.tech/tags/model.md>), [numbers](<https://devfeed.tech/tags/numbers.md>), [one](<https://devfeed.tech/tags/one.md>), [post](<https://devfeed.tech/tags/post.md>), [prompting](<https://devfeed.tech/tags/prompting.md>), [questions](<https://devfeed.tech/tags/questions.md>), [reliability](<https://devfeed.tech/tags/reliability.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [typesafe](<https://devfeed.tech/tags/typesafe.md>), [typesafe-jev](<https://devfeed.tech/tags/typesafe-jev.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

TypeSafe argues that Jev should supply judgments, not make decisions: reliability comes from composing those judgments in code. In the excerpt's separate proposal, Max Blade says harnesses should detect recurring tasks, suggest automations, track results, and modify workflows instead of limiting people to one human prompting one agent at a time. The excerpt gives no implementation or results for Blade's proposal, and says there is no evidence it works.

### Source excerpt

TypeSafe says reliable decisions are not something a model makes at all, they are composed in code out of Jev's judgments. Diogo Almeida says the opposite...

## Theo re-ran it: 5% fewer false reports, still bad

DevFeed: [Theo re-ran it: 5% fewer false reports, still bad](<https://devfeed.tech/articles/theo-re-ran-it-5-fewer-false-reports-still-bad-120949.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-05>)

Author: Jev Daily

Published: 2026-11-05T17:00:00Z

Content type: opinion

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [devstral](<https://devfeed.tech/topics/devstral.md>), [AI Architecture](<https://devfeed.tech/topics/ai-architecture.md>)

Tags: [acp](<https://devfeed.tech/tags/acp.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [cli](<https://devfeed.tech/tags/cli.md>), [company](<https://devfeed.tech/tags/company.md>), [git-worktrees](<https://devfeed.tech/tags/git-worktrees.md>), [inference](<https://devfeed.tech/tags/inference.md>), [model](<https://devfeed.tech/tags/model.md>), [parallel](<https://devfeed.tech/tags/parallel.md>), [reports](<https://devfeed.tech/tags/reports.md>), [stack](<https://devfeed.tech/tags/stack.md>), [team](<https://devfeed.tech/tags/team.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [theo](<https://devfeed.tech/tags/theo.md>), [voice](<https://devfeed.tech/tags/voice.md>)

### AI overview

Theo says rerunning the coding model through the CLI after fixing broken ACP bindings reduced false reports by about 5%. He still considers it unusable for real codebases, citing egregious hallucinations and describing it as far worse than a modest 5% shortfall. The excerpt also relays Greg Isenberg's separate account of a 50-plus-person, 12-country team that starts specs from voice notes, runs agents overnight in parallel git worktrees, and reviews version five.

### Source excerpt

Theo says fixing broken ACP bindings cut his false reports by about 5% and the model is still, in his words, utter garbage for real world work. Greg Isenberg...

## Prism beats Jev by two emails on a 200-email test

DevFeed: [Prism beats Jev by two emails on a 200-email test](<https://devfeed.tech/articles/prism-beats-jev-by-two-emails-on-a-200-email-test-118784.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-04>)

Author: Jev Daily

Published: 2026-11-04T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [decision-making](<https://devfeed.tech/topics/decision-making.md>), [email](<https://devfeed.tech/topics/email.md>), [Data analysis](<https://devfeed.tech/topics/data-analysis.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [beats](<https://devfeed.tech/tags/beats.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [decision](<https://devfeed.tech/tags/decision.md>), [decision-models](<https://devfeed.tech/tags/decision-models.md>), [email](<https://devfeed.tech/tags/email.md>), [jev](<https://devfeed.tech/tags/jev.md>), [margin](<https://devfeed.tech/tags/margin.md>), [models](<https://devfeed.tech/tags/models.md>), [sample](<https://devfeed.tech/tags/sample.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [test](<https://devfeed.tech/tags/test.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

In Arushi Gandhi's test of 200 of her own emails, Sarvam AI's Prism had the highest reported accuracy, classifying two more emails correctly than Jev. Jev was fastest and had relatively few false positives; OpenJev produced the fewest false positives but missed more cold emails. The excerpt does not provide model settings or broader evaluation details, so these results are limited to her test.

### Source excerpt

Arushi Gandhi ran 200 of her own emails through four decision models and says Jev came second by a margin of two emails, while being the fastest. On a sample...

## Jev picked login fields in 0.53s, Haiku took 7.83s

DevFeed: [Jev picked login fields in 0.53s, Haiku took 7.83s](<https://devfeed.tech/articles/jev-picked-login-fields-in-0-53s-haiku-took-7-83s-113289.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-02-w7>)

Author: Jev Weekly

Published: 2026-11-02T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Tool](<https://devfeed.tech/topics/tool.md>), [Claude](<https://devfeed.tech/topics/claude.md>)

Tags: [bias](<https://devfeed.tech/tags/bias.md>), [browser-tools](<https://devfeed.tech/tags/browser-tools.md>), [button](<https://devfeed.tech/tags/button.md>), [claude-haiku-5-5](<https://devfeed.tech/tags/claude-haiku-5-5.md>), [cost](<https://devfeed.tech/tags/cost.md>), [discount](<https://devfeed.tech/tags/discount.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [eval](<https://devfeed.tech/tags/eval.md>), [experiment](<https://devfeed.tech/tags/experiment.md>), [half](<https://devfeed.tech/tags/half.md>), [jev](<https://devfeed.tech/tags/jev.md>), [login](<https://devfeed.tech/tags/login.md>), [meta-ads](<https://devfeed.tech/tags/meta-ads.md>), [optimize](<https://devfeed.tech/tags/optimize.md>), [opus](<https://devfeed.tech/tags/opus.md>), [order](<https://devfeed.tech/tags/order.md>), [password](<https://devfeed.tech/tags/password.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Lightpanda reports that Jev selected the username, password, and submit fields in a login flow in 0.53 seconds with two calls, compared with Claude Haiku 5.5 using browser tools at 7.83 seconds and eight calls. The excerpt gives no page count or field-selection error rates, so it does not establish comparative accuracy. It also describes line-by-line search: use "nouls" for one expected dominant answer, or scores to highlight all relevant passages; processing each line or paragraph may make long documents costly. Theo built an evaluation from agents' actual tool use and failures, then iteratively revised tool descriptions across models; Opus initially failed more than half of over 80 cases. Separately, Mike Taylor warned Jev may favor the first-listed option as a judge, but published no evaluation or skew size, and the excerpt reports no counter-test.

### Source excerpt

Lightpanda clocked Jev at 0.53s and 2 calls against Claude Haiku 5.5 at 7.83s and 8 tool calls on the same login form task. The week's other lesson is...

## AI shoppers skipped the 50% off ad, Berman says

DevFeed: [AI shoppers skipped the 50% off ad, Berman says](<https://devfeed.tech/articles/ai-shoppers-skipped-the-50-off-ad-berman-says-111250.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-11-01>)

Author: Jev Daily

Published: 2026-11-01T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [consistency](<https://devfeed.tech/topics/consistency.md>), [synthetic monitoring](<https://devfeed.tech/topics/synthetic-monitoring.md>)

Tags: [ads](<https://devfeed.tech/tags/ads.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [black-friday](<https://devfeed.tech/tags/black-friday.md>), [bot](<https://devfeed.tech/tags/bot.md>), [cleanup](<https://devfeed.tech/tags/cleanup.md>), [discount](<https://devfeed.tech/tags/discount.md>), [effect](<https://devfeed.tech/tags/effect.md>), [examples](<https://devfeed.tech/tags/examples.md>), [jev](<https://devfeed.tech/tags/jev.md>), [meta](<https://devfeed.tech/tags/meta.md>), [meta-ads](<https://devfeed.tech/tags/meta-ads.md>), [metric](<https://devfeed.tech/tags/metric.md>), [pattern](<https://devfeed.tech/tags/pattern.md>), [rewrite](<https://devfeed.tech/tags/rewrite.md>), [run](<https://devfeed.tech/tags/run.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>), [tests](<https://devfeed.tech/tags/tests.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [writing](<https://devfeed.tech/tags/writing.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Matthew Berman says Jev evaluated 3,092 real Meta ads with 100 synthetic shoppers: among the top 100 picks, 2 had a 50% discount; in the panel, 32 chose a Black Friday ad with no discount and 12 chose the 50%-off ad. Aditya Umale says 5 of 9 recent writing-bot drafts went out unedited after a cleanup pass targeting patterns such as forced contrasts, clichéd closers, and phrases like "delve" and "game-changer." Greg Isenberg warns that agents rewarded for a metric may game it: an agent told to make tests pass could rewrite tests, while one told to close tickets faster could close them without solving the issue.

### Source excerpt

Matthew Berman's Jev agent ran 100 synthetic shoppers over 3,092 real Meta ads and says only 2 of the top 100 picks carried a 50% discount. If you are about...

## Shengkun Ye's people search runs at $0.0024 a lead

DevFeed: [Shengkun Ye's people search runs at $0.0024 a lead](<https://devfeed.tech/articles/shengkun-ye-s-people-search-runs-at-0-0024-a-lead-107211.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-29>)

Author: Jev Daily

Published: 2026-10-29T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [apollo](<https://devfeed.tech/tags/apollo.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [cowork](<https://devfeed.tech/tags/cowork.md>), [data](<https://devfeed.tech/tags/data.md>), [diogo-almeida](<https://devfeed.tech/tags/diogo-almeida.md>), [enrichment](<https://devfeed.tech/tags/enrichment.md>), [evals](<https://devfeed.tech/tags/evals.md>), [failure](<https://devfeed.tech/tags/failure.md>), [half](<https://devfeed.tech/tags/half.md>), [half-life](<https://devfeed.tech/tags/half-life.md>), [jev](<https://devfeed.tech/tags/jev.md>), [labs](<https://devfeed.tech/tags/labs.md>), [lead](<https://devfeed.tech/tags/lead.md>), [life](<https://devfeed.tech/tags/life.md>), [markup](<https://devfeed.tech/tags/markup.md>), [open](<https://devfeed.tech/tags/open.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [people](<https://devfeed.tech/tags/people.md>), [providers](<https://devfeed.tech/tags/providers.md>), [search](<https://devfeed.tech/tags/search.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Shengkun Ye's open-source Jev-powered people search runs leads through Apollo, Clay, People Data Labs, Hunter, ContactOut and Ploid in a waterfall, then pays for enrichment only for leads Jev keeps. Ye quotes a cost of $0.0024 per lead with no markup, compared with the $2,000 subscriptions it is intended to replace.

### Source excerpt

A second open source people search on Jev landed, this one quoting $0.0024 per lead by waterfalling six data providers and paying for enrichment only on the...

## Lightpanda: Jev picks login fields in 0.53s

DevFeed: [Lightpanda: Jev picks login fields in 0.53s](<https://devfeed.tech/articles/lightpanda-jev-picks-login-fields-in-0-53s-104327.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-28>)

Author: Jev Daily

Published: 2026-10-28T12:34:48.658127Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Agent Browser](<https://devfeed.tech/topics/agent-browser.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Claude](<https://devfeed.tech/topics/claude.md>)

Tags: [ads](<https://devfeed.tech/tags/ads.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [api](<https://devfeed.tech/tags/api.md>), [browser](<https://devfeed.tech/tags/browser.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [claude-haiku-5-5](<https://devfeed.tech/tags/claude-haiku-5-5.md>), [cost](<https://devfeed.tech/tags/cost.md>), [discount](<https://devfeed.tech/tags/discount.md>), [experiment](<https://devfeed.tech/tags/experiment.md>), [fields](<https://devfeed.tech/tags/fields.md>), [jev](<https://devfeed.tech/tags/jev.md>), [login](<https://devfeed.tech/tags/login.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [measurement](<https://devfeed.tech/tags/measurement.md>), [meta-ads](<https://devfeed.tech/tags/meta-ads.md>), [password](<https://devfeed.tech/tags/password.md>), [router](<https://devfeed.tech/tags/router.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [tools](<https://devfeed.tech/tags/tools.md>)

### AI overview

Lightpanda reports that Jev selected the username and password fields and submit button on a login page in 0.53 seconds using 2 calls. It compares this with 7.83 seconds and 8 browser-tool calls for Claude Haiku 5.5. The excerpt does not specify how many pages or sites were tested, so the figures do not establish performance across broader workloads.

### Source excerpt

Lightpanda put Jev on the login step of its browser agent and posted 0.53 seconds and 2 calls against 7.83 seconds and 8 tool calls for Claude Haiku 5.5. One...

## Sentry spends $263 a day on Jev for 10M events

DevFeed: [Sentry spends $263 a day on Jev for 10M events](<https://devfeed.tech/articles/sentry-spends-263-a-day-on-jev-for-10m-events-102468.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-27>)

Author: Jev Daily

Published: 2026-10-27T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>), [prompt](<https://devfeed.tech/topics/prompt.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [bench](<https://devfeed.tech/tags/bench.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [cloudflare](<https://devfeed.tech/tags/cloudflare.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [context](<https://devfeed.tech/tags/context.md>), [costs](<https://devfeed.tech/tags/costs.md>), [eval](<https://devfeed.tech/tags/eval.md>), [events](<https://devfeed.tech/tags/events.md>), [failure](<https://devfeed.tech/tags/failure.md>), [jev](<https://devfeed.tech/tags/jev.md>), [prompt-injection](<https://devfeed.tech/tags/prompt-injection.md>), [schemas](<https://devfeed.tech/tags/schemas.md>), [sentry](<https://devfeed.tech/tags/sentry.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [testing](<https://devfeed.tech/tags/testing.md>), [tools](<https://devfeed.tech/tags/tools.md>), [typesafe-ai](<https://devfeed.tech/tags/typesafe-ai.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Sentry runs Jev over 10 million events daily to screen for prompt injection before a coding agent reads an error report; Cloudflare Developers says this costs $263 per day. Sentry is testing an in-house classifier with the goal of matching Jev's accuracy at 90% lower cost. The excerpt does not report results or describe the classifier.

### Source excerpt

Sentry screens 10 million events a day through Jev at $263 a day to catch prompt injection, and is now testing its own classifier to hit the same accuracy for...

## Kush pulled Jev out of Puffle's harness

DevFeed: [Kush pulled Jev out of Puffle's harness](<https://devfeed.tech/articles/kush-pulled-jev-out-of-puffle-s-harness-100486.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-26-w7>)

Author: Jev Weekly

Published: 2026-10-26T14:17:35.931057Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [human review](<https://devfeed.tech/topics/human-review.md>), [benchmark overfitting machine learning](<https://devfeed.tech/topics/benchmark-overfitting-machine-learning.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [qwen](<https://devfeed.tech/topics/qwen.md>)

Tags: [about](<https://devfeed.tech/tags/about.md>), [actions](<https://devfeed.tech/tags/actions.md>), [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [all](<https://devfeed.tech/tags/all.md>), [also](<https://devfeed.tech/tags/also.md>), [also-on-the-timeline](<https://devfeed.tech/tags/also-on-the-timeline.md>), [article](<https://devfeed.tech/tags/article.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [best](<https://devfeed.tech/tags/best.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [false-negative](<https://devfeed.tech/tags/false-negative.md>), [file](<https://devfeed.tech/tags/file.md>), [flash](<https://devfeed.tech/tags/flash.md>), [harness](<https://devfeed.tech/tags/harness.md>), [jev](<https://devfeed.tech/tags/jev.md>), [making](<https://devfeed.tech/tags/making.md>), [qwen](<https://devfeed.tech/tags/qwen.md>), [reading](<https://devfeed.tech/tags/reading.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [worth-reading](<https://devfeed.tech/tags/worth-reading.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Kush removed Jev from the Puffle harness and returned to Qwen 3.7 flash after Jev repeatedly misjudged whether files were worth reading, leading the agent astray. Jev was intended to scout files and improve RAG precision, but Kush says false negatives--skipping relevant files--were especially costly and hurt their benchmarks. The excerpt also describes Lightpanda using Jev in judge() to classify page content, and Composio Firewall evaluating agent actions with Jev plus tuned SLMs; it does not explain how those uses performed.

### Source excerpt

Kush says Jev kept making wrong calls on whether a file was worth reading, so Puffle went back to Qwen 3.7 flash. Where a false negative is expensive, a cheap...

## TypeSafe discloses leak of all customer noul outputs

DevFeed: [TypeSafe discloses leak of all customer noul outputs](<https://devfeed.tech/articles/typesafe-discloses-leak-of-all-customer-noul-outputs-98898.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-24>)

Author: Jev Daily

Published: 2026-10-24T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [enterprise data protection](<https://devfeed.tech/topics/enterprise-data-protection.md>), [Security](<https://devfeed.tech/topics/security.md>), [Test-driven development](<https://devfeed.tech/topics/tdd.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [customer](<https://devfeed.tech/tags/customer.md>), [diogo-almeida](<https://devfeed.tech/tags/diogo-almeida.md>), [leak](<https://devfeed.tech/tags/leak.md>), [noul](<https://devfeed.tech/tags/noul.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [typesafe](<https://devfeed.tech/tags/typesafe.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

TypeSafe AI disclosed that it became aware at about 6:49 p.m. Friday of an unauthorized leak it described as containing all customer "noul" outputs. A user, Jarrod Watts, replied that his data was included, and Diogo Almeida apologized, saying the company needed to do better. The excerpt provides no further details about the leak or the company's response.

### Source excerpt

Diogo Almeida says TypeSafe AI became aware of a massive unauthorized leak of all customer noul outputs at 6:49PM Friday, and at least one builder has already...

## Composio puts Jev in front of every agent action

DevFeed: [Composio puts Jev in front of every agent action](<https://devfeed.tech/articles/composio-puts-jev-in-front-of-every-agent-action-97797.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-23>)

Author: Jev Daily

Published: 2026-10-23T17:00:00Z

Content type: release

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [Claude](<https://devfeed.tech/topics/claude.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [composio](<https://devfeed.tech/tags/composio.md>), [decision-models](<https://devfeed.tech/tags/decision-models.md>), [firewall](<https://devfeed.tech/tags/firewall.md>), [harness](<https://devfeed.tech/tags/harness.md>), [information](<https://devfeed.tech/tags/information.md>), [jev](<https://devfeed.tech/tags/jev.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [models](<https://devfeed.tech/tags/models.md>), [prompt-injection](<https://devfeed.tech/tags/prompt-injection.md>), [routing](<https://devfeed.tech/tags/routing.md>), [simple](<https://devfeed.tech/tags/simple.md>), [small](<https://devfeed.tech/tags/small.md>), [small-language-models](<https://devfeed.tech/tags/small-language-models.md>), [small-models](<https://devfeed.tech/tags/small-models.md>), [sydney](<https://devfeed.tech/tags/sydney.md>), [talk](<https://devfeed.tech/tags/talk.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Composio Firewall evaluates agent actions before execution using Jev and small language models tuned on Composio usage data. It checks both outgoing actions and information returned through Composio, aiming to catch prompt injection, and is live for all Composio developers. Separately, Sanity built an internal harness for employees to create bespoke agents, with Jev handling routing; its builder reports that it has worked pretty well so far. Sydney Runkle's advice on agent decision models was to start with simple, inexpensive models and use an advanced model only when the task needs one.

### Source excerpt

Composio now runs every action an agent takes through Jev and its own tuned small models before it executes, and published no latency or cost figures for the...

## Lightpanda put Jev in its binary for page triage

DevFeed: [Lightpanda put Jev in its binary for page triage](<https://devfeed.tech/articles/lightpanda-put-jev-in-its-binary-for-page-triage-95744.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-22>)

Author: Jev Daily

Published: 2026-10-22T14:22:18.019331Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [implementation](<https://devfeed.tech/topics/implementation.md>), [static linking](<https://devfeed.tech/topics/static-linking.md>), [API Business Models](<https://devfeed.tech/topics/api-business-models.md>)

Tags: [article](<https://devfeed.tech/tags/article.md>), [automation](<https://devfeed.tech/tags/automation.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [content](<https://devfeed.tech/tags/content.md>), [cost](<https://devfeed.tech/tags/cost.md>), [demo](<https://devfeed.tech/tags/demo.md>), [diogo-almeida](<https://devfeed.tech/tags/diogo-almeida.md>), [docs](<https://devfeed.tech/tags/docs.md>), [faster](<https://devfeed.tech/tags/faster.md>), [inside](<https://devfeed.tech/tags/inside.md>), [integration](<https://devfeed.tech/tags/integration.md>), [jev](<https://devfeed.tech/tags/jev.md>), [latency](<https://devfeed.tech/tags/latency.md>), [llm](<https://devfeed.tech/tags/llm.md>), [models](<https://devfeed.tech/tags/models.md>), [order](<https://devfeed.tech/tags/order.md>), [pitching](<https://devfeed.tech/tags/pitching.md>), [product](<https://devfeed.tech/tags/product.md>), [questions](<https://devfeed.tech/tags/questions.md>), [real-world](<https://devfeed.tech/tags/real-world.md>), [run](<https://devfeed.tech/tags/run.md>)

### AI overview

Lightpanda embedded Jev in its binary and built judge(), which answers predefined questions to determine whether a page has real content and classify it as an article, listing, product page, or docs. The excerpt reports one call takes about 0.2 seconds; a demo covered 30 sites in roughly 14 seconds. Separately, UiPath added a TypeSafe connector to its Integration Service for workflow decisions. TypeSafe says Jev is about an order of magnitude cheaper and faster than an LLM judge; the excerpt attributes the size of that gain to TypeSafe itself.

### Source excerpt

Lightpanda compiled Jev into the Lightpanda binary and built judge() on it, one call at roughly 0.2s to tell a real article from a cookie wall. Separately...

## Mike Taylor: Gemini beat stacked Jev on image detection

DevFeed: [Mike Taylor: Gemini beat stacked Jev on image detection](<https://devfeed.tech/articles/mike-taylor-gemini-beat-stacked-jev-on-image-detection-92799.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-21>)

Author: Jev Daily

Published: 2026-10-21T13:51:57.780555Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [image recognition](<https://devfeed.tech/topics/image-recognition.md>), [Image](<https://devfeed.tech/topics/image.md>)

Tags: [better](<https://devfeed.tech/tags/better.md>), [call](<https://devfeed.tech/tags/call.md>), [detection](<https://devfeed.tech/tags/detection.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [image](<https://devfeed.tech/tags/image.md>), [jev](<https://devfeed.tech/tags/jev.md>), [less](<https://devfeed.tech/tags/less.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Mike Taylor says Gemini found bounding boxes in an image better and for less money than Nathan LeClaire's approach of stacking Jev on itself to divide the image, because Gemini needed one call. The excerpt gives no measurements or further comparison details. It also describes a separate production-evaluation loop: calibrate a judge against human labels, flag low-scoring outputs, have critics diagnose failures, add accepted repairs to training data and evals, then fix and rerun.

### Source excerpt

Nathan LeClaire stacked Jev on itself to find bounding boxes in an image, and Mike Taylor says Gemini did a better job in one call, for less money...

## Zachi: Cognition's Devin Fusion chart is "a crime"

DevFeed: [Zachi: Cognition's Devin Fusion chart is "a crime"](<https://devfeed.tech/articles/zachi-cognition-s-devin-fusion-chart-is-a-crime-90976.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-20>)

Author: Jev Daily

Published: 2026-10-20T17:00:00Z

Content type: comparison

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [Claude](<https://devfeed.tech/topics/claude.md>), [anthropic](<https://devfeed.tech/topics/anthropic.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [artificial-analysis](<https://devfeed.tech/tags/artificial-analysis.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [code](<https://devfeed.tech/tags/code.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [context](<https://devfeed.tech/tags/context.md>), [cost](<https://devfeed.tech/tags/cost.md>), [devin](<https://devfeed.tech/tags/devin.md>), [domain](<https://devfeed.tech/tags/domain.md>), [fusion](<https://devfeed.tech/tags/fusion.md>), [gap](<https://devfeed.tech/tags/gap.md>), [gpt](<https://devfeed.tech/tags/gpt.md>), [gpt-6](<https://devfeed.tech/tags/gpt-6.md>), [gpt-6-luna](<https://devfeed.tech/tags/gpt-6-luna.md>), [harness](<https://devfeed.tech/tags/harness.md>), [index](<https://devfeed.tech/tags/index.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [opus](<https://devfeed.tech/tags/opus.md>), [opus-5-5](<https://devfeed.tech/tags/opus-5-5.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>), [writing-criteria](<https://devfeed.tech/tags/writing-criteria.md>)

### AI overview

Cognition says Devin Fusion scored 68 on the Artificial Analysis Coding Agent Index and cost 31% less per task than Claude Code with Opus 5.5, describing a setup with an Opus 5.5 lead and GPT-6 Luna sidekick. Zachi calls the comparison chart misleading because it omits more than 30 models priced below $5; including them, Zachi says, would shift the attractive quadrant and narrow the apparent gap. The excerpt also reports Sydney Runkle's ordering of agent improvements: optimize context first, then the harness, then the model weights.

### Source excerpt

Cognition says Devin Fusion scores 68 on the Coding Agent Index at 31% lower cost per task than Claude Code with Opus 5.5. Zachi says the chart behind it...

[Next page](<https://devfeed.tech/sources/jev-daily.md?cursor=WyIyMDI2LTEwLTIwVDE3OjAwOjAwKzAwOjAwIiwgIjI4MjRkZjUyLWRkOTQtNDA0My1hMTBkLWYzNmIwMTJiNWI3YSJd>)