Alessandro Bahgat

Built this Scope 3 prototype (Scopewise) over the weekend for a distributed #EarthDay hackathon nudging agents from my phone between family activities. If one parent with fragments of weekend time can get here, what's a full sustainability team's excuse for not even prototyping?

@gabor Others suggested Playwright. I'm a fan of Chrome Dev Tools MCP. I have specs and requirements checked in and have agents test them regularly in the browser (like a human QA agent). And if they catch regressions, I have them write and commit unit or integration tests.

How do you let the models test the apps they vibecoded? Like click through the UI, sign in as different test users, etc. - what's the state of the art here?

I spent a few hours over the last couple of weekends getting local AI models to play Zork, a 40-years old text adventure. On paper, another toy project. In practice, a great lesson on agent evals. abahgat.com/blog/same-agen…

AI can now pull data from your issue tracker, error monitoring, Slack, and codebase: but the final deliverable is still a manually assembled slide deck that no AI can edit properly. The last mile of engineering leadership is copy-paste.

You know it's 2026 when building a B2C app takes less time than summarizing what you built in Google Slides.

Congrats to our top authors this week! 🏆 @pcescato, @rsdgls, @abahgat, @shrsv23, @nicolas_frankel, @morgantwillis, and @aryanc193. This week's lineup features insights on taking AI agents from POC to production, the "ghosts" inside training datasets, and much more. dev.to/devteam/top-7-…

I did a deep dive into frequency bias, why "Zero-Prompt Tests" are becoming essential, and how to keep your agents.md from becoming a mountain of legacy cruft: abahgat.com/blog/the-ghost…

We’re increasingly moving from managing code to managing the attention and memory of our agents. To keep this from becoming "Contextual Debt," you have to be as ruthless about pruning your instructions as you are about pruning your codebase.

I’ve been working with MCP recently and noticed something surprising. Even when models know a new standard exists, they are often haunted by the statistical momentum of their own training data. I call it "The Ghost in the Training Set." A few thoughts on managing this below.

@GergelyOrosz I think it would be a very interesting story to cover what good "Internal tests and benchmarks" look like and how companies are evaluating these tools.

From a dev at a large tech company: “We were only allowed to use GitHub Copilot as an AI IDE. It was OK. But then more and more of us used Cursor on side projects and it was *so much better* Luckily we have have a dev platform team and we told them we want to use Cursor. So they ran these internal tests and benchmarks and found that it worked a lot better. They now sorted everything and we can all officially use Cursor - and it’s been such a big positive change!”

✈️🌎 We're teaming up with @Google as the first company to use the Travel Impact Model (TIM) to calculate past emissions from our business travel. Discover more 👉 bit.ly/3TDkJvj

@gabor Right? ;)

Is there an AI assistant that observes my actions on the computer, and then offers to automate my workflows when it has learned them? Feels like a product that is now possible.

Extremely proud to be part of an incredible engineering team at #Google playing a role in developing a groundbreaking tool for environmental transparency in air travel, in collaboration with #EASA. Together, we strive for a greener future.

Whiteboards are, at the same time 1. among the most compelling reasons for returning to the office 2. among the biggest obstacles to inclusive meetings for hybrid teams

Want to start a startup? A $925 million market just appeared.

@saarw @GergelyOrosz I've seen this too. Sometimes it's a case of "second-system effect": having seen the cost of poor engineering practices in the past leads to over-investing in them in the future. One of the parts of The Mythical Man-Month that I still see today en.m.wikipedia.org/wiki/Second-sy…

@GergelyOrosz It's a balance. Easy for teams to focus too much on form rather than delivering value Mid-experienced people (around 5 years) run the greatest risk of being overly concerned with form. Perhaps fear of stepping outside a small domain they've mastered keeps them polishing it

“I joined Well Known Big Tech and I’m depressed at the engineering culture of my team. I expected SO much more in quality, best practices, hygiene….” If you’ve been in these shoes, what rational reasons have you seen for it? Any advice? (Happens more frequent than many think!)

@GergelyOrosz Blind feels like "4chan for work" sometimes. ;)

“I’m a Senior eng at {large tech company} in London. Do you think I could get a Staff position at another, similar company?” “Do you know what Staff engineers do at other companies?” “Kind of… I read on Blind.” “Have you talked to any?” “No.” “So here is my advice: (cont’d)

@rainbowbreeze Ho fatto proprio fatica a leggerlo. A parte il contenuto che non condivido, trovo lo stile in cui è scritto troppo pesante e inefficace. Vorrei che qualcuno facesse un equivalente Italiano di plainenglish.co.uk/how-to-write-i…

In un comunicato "istituzionale", c'è tutto quello che server per capire come mai la PA italiana ha fallito totalmente nell'abbracciare un modello di lavoro più flessibile, in questi 18 mesi di pandemia, e negli anni precedenti. funzionepubblica.gov.it/articolo/minis…