Dor Blech retweeted
Security Center 2.0 is live in @Base44 !
timing couldn't be more interesting, but we've been building Base44 to be the most secured vibe coding platform for a while now.
We built the Security Center a few months ago, and this week we released phase 2
Phase 1 gave workspace admins one view of every app in the company - who owns it, who it's shared with, what's published, what the open risks are.
Phase 2 makes it work on its own:
- Scheduled scans: weekly for your most critical apps and monthly for the whole workspace, with a history of every run (Enterprise)
- Notify owner: one click and the app owner + collaborators get an email and in-app summary of the open risks, so admins don't need to chase anyone
- Way more visibility: what each app connects to, which AI providers it uses, stored secrets, credit spend and more. Everything filters and exports
When anyone in a company can build an app, admins need to know what's out there and that it keeps getting checked.
It's live now under Workspace Settings → Security Center
Dor Blech retweeted
Excited to back @WoodingtonBen , @Elise__Jenkins and the great @coherenceneuro team
Alongside @foundersfund, @OpenAI and other fantastic investors.
Coherence Neuro is building an implant that treats cancer while learning and adapting.
Shortening treatment iterations - collapsing the time it takes to measure the disease, learn, adapt and treat - will prove out to be key in effectively controlling cancer.
I’m extremely excited about their approach.
They’re hiring roles across SF and Cambridge UK.
If you want to work on hard problems that truly matter - check out their careers page.
Dor Blech retweeted
We get to see many beautiful use cases at @Base44 that make us wake up every day with a feeling that we’re doing something that is very net positive in the world.
For me - this is one of the top ones -
Dotan Haas Hershkovitz built “Check it out” - after losing her mother to cancer - to help women get checked for breast cancer.
Some great stuff inside:
- Find a doctor that fits you, by location and how they approach care
- WhatsApp or email reminders to get checked once or twice a year
- Matches you with another woman - if you don't want to go to the appointment alone
- Info on self exams and a risk questionnaire
Since publishing the app she has had thousands of women using it - getting checked, helping each other with info and going to checkups together.
Some of them found breast cancer early because of it. Which is hard to overstate how important that is for effective treatment.
I love every part of this story.
We will do our best to support Dotan, and other builders building for good.
Dor Blech retweeted
שלום לכולם שמי רועי זגה, סטודנט לביולוגיה וכימיה באוניברסיטת תל אביב. אני מציע שיעורים פרטיים במחירים שפויים בביולוגיה, כימיה, מתמטיקה ולשון באיזור רמת גן. בעל ניסיון של יותר מ-8 שנים בתחום עם עשרות תלמידים במקצועות שונים וגילאים מגוונים כולל 4 שנים כחונך במרכז למידה של עמותת מסלולים.
מוזמנים ומוזמנות לפנות אליי לתיאום שיעור ♥️
Dor Blech retweeted
We got a new careers page
Check it out, see how pretty our team is when we force them to look like they’re having fun, and fill in your contact to join us on this adventure.
some cool roles are open world wide - Head of Security, AI Engineer, Director of Sales Development, Influencer Marketing Lead and more.
link in the first comment.
Dor Blech retweeted
What an unbelievably dumb take
Martin Shkreli says AI will not deliver clinically approved drugs anytime soon.
“I don't think there'll be a single AI-relevant drug in the next 10 years. AI does not help drug discovery. If you've ever built drugs, and raise your hand if you've made a drug, gotten FDA approval for a drug... It's pretty hard to do. And we've had computers for a long time. I built a Cray cluster to analyze drugs. Now I can use, you know, GPUs, it's great, but there isn't much AI can help us…”
Silicon Valley loves to pitch LLMs as a biological cheat code, but real drug discovery is a wet-lab nightmare that raw token prediction just doesn't solve.
@theallinpod
Dor Blech retweeted
Some cool trends we’re seeing for “vibe coding” (starting to hate the name) -
General:
- Apps are becoming more complex - when looking at the 20th prompt - apps now have 1.9x lines of code than a year ago
- 27% of the conversations/prompts people send to @Base44 are now around marketing / strategy, instead of app building
Organizations:
- Enterprises are adopting app building much faster than the pace consumers are building (6 months ago it was the opposite). Also an avg enterprise user is building 2.7x more apps than a consumer user.
- Most companies start by building niche use cases (e.g. lead dashboard for a campaign) that over time are crunching into core systems (CRMs, ticketing systems)
- BI and custom data tools are growing bigger
E-commerce:
- Users are now building “ephemeral” e-commerces stores for specific trends that generate revenue and close shop after a few months. Ephemeral e-com stores are up 3.6x than 4 months ago (also has to do with improvements in our payments system onboarding)
- drop-shipping is down (0.7x vs 4 months ago)
Dor Blech retweeted
בחיים לא הייתי צריכה את העזרה שלכם כמו עכשיו!
אני לא בארץ ו ומילקה ברחה מבית של חברה באיזור נוף ים בהרצליה.
בבקשה בבקשה תפיצו ותעזרו למצוא אותה אני מתחננת 🙏🏽❤️
Dor Blech retweeted
500 K-12 students spent the weekend in Austin building apps. @HCforA put the whole thing together.
31 teams submitted. A few cool stuff developed there:
A navigation for blind people, paired with smart glasses and a cane.
An app for family caregivers to claim the benefits their aging parents already qualify for.
A real-time ASL translation in both directions.
all from such young kids, that's wild!
I'm proud of @Base44's education team.
Dor Blech retweeted
Tl;Dr - Base 1 won
We built the same three apps with five models on @Base44 and ran Tenzai's AI hacker against all 15.
Tenzai found 254 vulnerabilities, and while every model had flaws, only one had zero high or critical findings - and was the only model that never exposed sensitive data to unauthenticated users.
*Model choice is a security decision*
Full breakdown: tenzai.com/blog/five-models-…
ALT Five Models Walk Into an App Builder. Security Gets Interesting.
Dor Blech retweeted
TL;DR:
We're looking for exceptional engineers / data scientists
who are passionate about the impact AI can have on cancer drug discovery, and want to make an impact on how cancer is treated.
---
2nd in my series of cancer investments:
Starget Pharma
Starget is probably the most AI-pilled company I’ve seen in the bio space,
which helps them move at an unbelievable speed, especially considering they’re developing cancer drugs.
Starget is developing next-gen radioligand therapies -
a modality I believe will eventually play a substantial role in managing a decent portion of cancers.
Why:
- Imaging first - being able to see which tumors the targeting agent reaches before starting treatment is a very powerful advantage.
- Radiation already works - radioligands allow us to deliver treatment systemically, including to metastases.
- Separating targeting from cancer cell killing (targets are not necessarily pathways to block).
- I think much of the upside is still ahead: better targets, better delivery, smarter dosing, and combinations.
Starget is quite unique - it’s building much of the stack in-house,
from AI-led molecule design in silico to closing the loop with lab results.
We’ve already dosed the first patient and have a few programs running (which I’m very optimistic about).
And we’re now looking to scale the engineering and AI team with people who are passionate about:
-Building systems to identify novel therapeutic targets.
- Running millions of proteins through scoring models at scale.
- Building generative molecule design processes.
- Closing loops between models and the wet lab.
It ain’t going to be easy, but def going to be fullfilling.
If you know anyone - tag them!!
or have them send a few lines about themselves and what they’ve built to the email in the comments
Dor Blech retweeted
Prediction: we'll soon stop caring about PRs
We used to talk in lines of code, then commits, then PRs - always rising in abstraction. Every time the old unit got cheap, we moved up a level. Agents are making PRs cheap now.
Long-running coding sessions will become the new main unit of work - from idea, up until real resolution.
Everything should, and will, happen in the same session, until it's done:
- a bug fix confirmed in production
- an A/B test concluded and the winner merged
- a large feature, with several iterations, from idea to real business impact
When coding agents started to become popular 1.5 years ago, their main purpose was to author a PR. Humans took care of the reviews, the fixes and the deployment.
Next quickly came the various "fix PR comments" skills. Anthropic even built that into Claude Code.
The next phase was verification - getting the agent's feedback loop on a single PR to be as autonomous as possible. I guess most companies nowadays are either here or before that.
Now we're working really hard to find ways to auto-approve most PRs in a safe way. I think most cutting-edge companies are going in that direction too.
Each step pushes the human handoff a bit later. Once most PRs are auto-approved, there's no real reason for the session to end at the PR. And the work doesn't end after a PR is deployed:
- there's usually some kind of production monitoring to do
- sometimes a feature flag needs to be gradually rolled out
- and then comes iterating and fixing accordingly
We tried doing that using an external agent monitoring custom production metrics per PR, after one is merged. But it's cumbersome - you need to feed it all the context again, and if it finds an issue, you're back to square one: a new coding session, or manually going back to your previous one and pasting the findings. Not ideal.
Turns out agents have the same problem human teams have - every handoff loses context.
So my prediction is that we'll start treating the session (or "task", whatever we end up calling it) as the new main unit of work. Opening PRs, deployments, feature flag rollouts, reacting to user feedback and iterating - all in a single session that's only closed once its goal is achieved. Big or small.
Half a year ago, taking a big feature from idea to verified impact took dozens or hundreds of human decisions. Models getting smarter are cutting this number - coding better from the beginning, with better tools, better automatic reviews and verification. I believe we're not far from people just becoming directors of a long list of sessions, making a human decision once in a while.
Bonus prediction - each finished session will be the point where the collective brain of the company gets recomputed: agents.md, best practices, skills, etc. Goal based dreaming.
Dor Blech retweeted
with the latest model releases it seems that the value of the instructions / system prompt part of the harness is going to zero.
Models need less instrcutions-type steering to deliver results.
Remember when everyone talked about the rise of prompt engineers...
Also I feel nostalgic for the weeks I spent writing instruction poems to get claude sonnet 3.5 to output workable code in @Base44
Dor Blech retweeted
We're building more and more towards agents using our platform, not only humans,
allowing our users to plug the product into their agent and run their business end to end via whatever they're using.
It's such an interesting new way of building and it raises a lot of weird questions
do we build agents that speaks to our @Base44 agent? or do we let external agents write code into base44?
do we just open up all APIs (SEO, security scans, etc)? does it comes with a list of skills on how to use base44 or is that too much of an onboarding?
how do we benchmark this whole thing?
