Blog
Thoughts on AI browser agents, browser automation, and building dassi.
-
CRM CSV import or browser updates: choose by the work you have
Decide whether prepared prospect data belongs in a CRM CSV import or a short list of reviewed record edits. Includes a decision table, HubSpot matching and rollback rules, and an exception sheet for conflicting values.
-
Qwen3.8 27B Scores 17.8 in Browser Use's Harness and 49.7 in dassi on BU Bench V2
dassi on the open-weight Qwen3.8 27B scored 49.7 on BU Bench V2's public 55-task subset and fully completed 27 of 40 sampled Odysseys tasks. It is slow, though: 44 minutes a task, and 40% of BU Bench tasks hit our 60-minute limit.
-
Shopify bulk editor, CSV or a browser assistant: which fits your update?
Choose between Shopify's bulk editor and CSV import for a product update, and see when the real work is gathering supplier data first. Includes a decision table, CSV overwrite checks and a read-only preparation prompt.
-
Compare supplier price lists by SKU and flag changes
Compare an old and new supplier price list by SKU and get a reviewable change report before you touch store prices. Includes an illustrative delta table, unit and currency checks, a reusable prompt and a pre-update checklist.
-
Check service prices, durations and booking links across your website, booking page and listings
Build an exception report that compares each service on your website, Acuity booking page and Google Business Profile against an owner-approved menu. Includes an illustrative comparison table, a reusable read-only prompt and a review checklist.
-
Turn prospect research into a deduplicated CRM import
Build an evidence-linked prospect table and check it against your CRM export before import. Includes an illustrative review table, HubSpot matching rules, a reusable prompt and a pre-import checklist.
-
Match supplier photos to Shopify variants before updating products
Build a reviewed image-to-variant mapping from SKUs, source filenames and supplier labels before you touch Shopify. Includes an illustrative mapping table, a reusable prompt and a pre-update checklist.
-
Move Your Service Menu Into Acuity and Check Every Field
A field-by-field checklist for moving services into Acuity Scheduling: names, durations, prices, currency, calendars, locations and booking links, with a comparison table and a review prompt.
-
Build a candidate brief with a source for every claim
A recruiter candidate summary template where every fact points to a resume line or approved profile, gaps are marked, and nothing is ranked or inferred. Includes a fictional example and a reusable prompt.
-
Anthropic Is Paying Akamai $11.6B for the Edge. Your Browser Agent Already Lives on One.
Anthropic's $11.6B Akamai deal buys proximity to users. For a browser agent, the closest edge is the logged-in Chrome tab you already have open on your desk.
-
Turn a Supplier Catalog Into a Checked SKU Spreadsheet
Convert supplier product pages into a source-traced SKU table before any store update. Includes real supplier screenshots, a reusable prompt and a verification checklist.
-
dassi on DeepSeek V4.1 Flash Scores 79 on BU Bench V2 for 8 Cents a Task
Graded with Browser Use's own judge code, dassi running DeepSeek V4.1 Flash scored 79.0 on the public 55-task BU Bench V2 subset, at a measured $0.08 in model calls per task. Browser Use's own run of the same model at the same reasoning level scored 41.5.
-
dassi on Gemini 3.8 Flash Scores 57 on BU Bench V2, and a Model 12 Times Cheaper Beats It
Same agent, same 55 BU Bench V2 tasks, same judge: dassi on Gemini 3.8 Flash scored 56.9 at $1.01 a task, and on DeepSeek V4.1 Flash it scored 79.0 at $0.08. Our earlier Gemini number was wrong, and this replaces it.
-
dassi Scores 85% on Odysseys With the Official Scorer, on a Flash Model
Graded with the Odysseys authors' own scorer and judge, dassi on Gemini 3.8 Flash fully completed 170 of 200 long-horizon web tasks. Our first number, 91.5%, came from our own judge; here is why the two differ.
-
Job Postings Show AI Automation Hitting Logged-In Browser Work First
The first jobs AI automation is shrinking live in logged-in browser tabs. Three chores from those listings, run in your own Chrome, plus where it breaks.
-
Meta's Muse vs an AI Browser Agent: Only One Is Logged In to Your Work
Meta put its Muse agent in a wearable and camera-free glasses. None are signed in to your work. An AI browser agent in your logged-in Chrome already is.
-
Ema Raised $77M for AI Employees. Each One Still Needs a Connector to Reach Your Apps.
Ema raised $77M for AI employees, but each one needs connectors and service accounts. Comparing three enterprise tasks against a browser agent in your SSO tab.
-
Ron Johnson Doesn't Buy AI Shopping. The Real Problem Is Your Agent Checks Out as a Stranger.
Ron Johnson says AI shopping won't work. He's half right: AI shopping agents boot a blank cloud browser with no cart, no addresses, no loyalty tier, no history.
-
Google's $899 Googlebook Runs Gemini Beside Your Tabs, Not Inside Them
Google's $899 Googlebook is a browser in a shell, and Gemini still can't see the Salesforce tab you're logged into. Why an ai browser agent beats new hardware.
-
What a Chrome Extension AI Agent Can and Can't See
A concrete map of what a side-panel browser agent reads, what it can't touch, and where your data goes when you bring your own model key.