Been building AI for 20 yrs. Refactoring all of it Can be found near @sirblueborg and/or a beach ex-CEO, Founder @madstreetden @vue_ai
Joined August 2013
- Tweets21.5K
- Following1K
- Followers22.9K
- Likes28.8K
A constant reminder that we're in the ugly quick & dirty MVP phase of AI. We didn't fix the bloated SI bench, we just gave it API keys and turned the agent bench into a bloated one.
"Throw an agent at it" is the new "throw headcount at it."
With all the drama going on with all things compute, security & the economy in general, I wager 2027 is the year we're going to start seeing themes that counter 'growth at all cost' arc of '25-'26.
Touchgrass. Seegrass. Smellgrass.
Undoing effects of 18 hrs a day with claude & codex by making it my 18/7 backdrop!
3
20
1,831
This is SUCH a great time for designers, design, agencies, branding, storytelling ...
1. We're going to see a complete explosion of these services, new age, new skills, entirely new aesthetic vs. the same old boring blue/purple SaaS age.
2. Designers, design & mktg teams are not being let go because there's no work for them anymore. Its because the bar for what qualifies as a solid team has changed fundamentally. We're going to need SOOOOO MANY MORE of these teams in the post AI era. It's a great time to double down, ramp up those new skills and deliver outsized returns for companies
On that note - twitter - if anyone knows some fabulous AI-pilled designers and brand storytellers - point them to me?
“Excellence is a moving target and we will keep changing” is the singular line to takeaway from that Sequoia memo.
And to adapt a gazillion times faster than what we did as humans is a Q of focus. Which is a horror story given how distracted all of us are thx to our AI conquests! Cracking this is everything.
1
3
1,419
Somebody make me stop. You can skin this cat a 100 ways. So many metaphors.
Sleep consolidation. The brain consolidates what it learned during sleep, not while learning. AI builds need deliberate pauses, iterations, strategy to consolidate.
1
5
1,251
As someone coming back from a couple of years of total exhaustion, and coming back with full energy very conscious of the healing process, the parallels to the AI madness and the impending *burn are not lost on me. Time for a ton of writing on governance and what follows the madness that is our life right now :)
Anyone who's had a deep cut knows this. The body closes the wound fast, in days. But what it closes it with is scar tissue. Strong enough to hold but never as strong as what was there before. The real healing happens slowly, underneath, over months.
Everyone is building AI at wound-closing speeds :) Just don't mistake the scar for the skin.
ashwini asokan retweeted
Opus 5.5 Medium + Blender MCP gets more interesting when you look at how the work is assigned.
A developer building a medieval game found that handing Claude the whole project left everything feeling 70% finished, followed by constant tweaking.
What worked better:
• A clear task list
• Small, focused tasks
• One task at a time
They report a major improvement in output quality.
How much of an agent’s performance depends on where we draw the task boundaries?
Perspective is the real currency in a post AI era.
We're all pretty shitfaced right now, overloading our days with tools, speed, volume of everything that it really is beginning to feel like everything everywhere all at once.
Find time and more importantly the discipline, to step away from the matrix y'all. Your AI-pilled lives are only as effective as your ability to create meaningful outcomes with them.
ashwini asokan retweeted
Very timely note!
I've noticed AI coding is 10–20x for the first month or 2 of a project. Then complexity kicks in, and I either ship brittle code or slow down to really understand what’s being built. The speedup slows down fast.
We’ve seen this before. In the steam era, a new source of power dropped into an old way of working changed surprisingly little. The gains came when the work itself was reorganized around it. Also true of Intel and how the Silicon guys built their foundries. Many similar examples.
Our old way of working runs everything through one human’s understanding. Every change still passes through someone’s head. AI made writing code cheap but left that bottleneck untouched.
The thing is, software already knows the answer. Microservices were kept small so their contracts and specs stayed well contained. Then, to make AI coding work, we all moved to monorepos so agents could see everything, and we started shipping bigger, more tangled software.
Speed comes from isolation. Growth should mean adding more well-specified components, not making one system larger. You can do this inside a monorepo, but only if you compartmentalize hard enough that nobody, human or agent, has to do the cognitive switching across the whole thing.
Build so each part can be verified locally, against its contract, instead of in one head. Seems like this could make speedup sustainable, a possible antidote to the poison pill.
All of us understand that useful software is getting easier to create. But generating code, completing end-to-end engineering work, and creating business value are moving at very different rates.
Digging into this from different angles. Epoch, for example, estimates that a fixed level of AI performance is getting cheaper 13x per year for the past 4 years. That's across math, science, and game projects.
Anthropic's 100k C compiler experiment is well known: 20k API spend, roughly $200 per kLOC of code, but the compiler had major limitations, lots of bugs, and wasn't really comparable to a fully functional C compiler.
An early 2026 Microsoft study estimates 25% more merged pull requests. Anthropic has shared bigger stats with us. What's not clear, though, is whether this translated into business value. It's an observational study, and in general, I am skeptical of observational data about increased activity.
A more recent September 2026 paper has impressive stats: 240%+ more commits, 80% more projects, 30% more releases, but it's output gains, not cost reductions.
It's clear that cheaper software creation is going to make previously uneconomical software worth building: bespoke tools, small markets. I built a bunch of app & skills I would never have bothered to write actual code for before, for example. why? because it's cheap and easy.
Estimating value for these can be tricky. If, for example, somebody never created a piece of software that could create 10k of value because software costs 50k to create, and all of a sudden that 50k became 1k, the total economic value that's created is 9k, not 49k. This kind of "more is more" can be tricky to measure in terms of actual gain, either for a company or for society.
The one other thing that people don't seem to keep in mind is that AI created software (all software?) comes with a poison pill: catastrophic bugs.
If our ability to create new software grows out of proportion with our ability to detect them or, worse, fix them, we end up in a crisis-driven world.
The world of crypto offers a warning here: there's a mind-numbing repetition of bugs with massive impact. Think about that. Would you tolerate that from your bank?
Thinking about this as well as value capture from other world-changing inventions. Comment with ideas and thoughts.
The sheer scale of output we're producing with AI now ... seeing significant issues with memory, ability to recall what I did even a few days ago.
Making Claude draw up an infographic of the work I did every week, the artifacts I produced etc retrospectively to serve as continuous reminder.
9
19
3,665
ashwini asokan retweeted
🚨Breaking news: US. banks are taking hits from AI agents
“findings have alarmed banking executives, who are racing to restrict access to their systems as AI moves at a velocity faster than financial institutions can respond”
Predictions:
1. No matter how much chaos generative AI causes, the AI industry will not take real responsibility
2. The Trump administration will not hold them responsible
SCOOP: Major U.S. banks are detecting unauthorized third-party AI agents inside customer accounts, including some attempting to take actions that appear to be independent of user instructions, sources told Axios.
The findings have alarmed banking executives, who are racing to restrict access to their systems as AI moves at a velocity faster than financial institutions can respond, multiple industry and bank executives told Axios.
Several banks are now preparing for worst-case and even “day after” scenarios, including how to rapidly secure or restore customer accounts if things go wrong, the people said.
Several bank executives told Axios they're struggling to distinguish between human users and AI agents. One executive said banks also struggle to determine whether agents are authorized by customers or potentially operating without their permission, more akin to rogue agent behavior.
Read my latest for Axios here: axios.com/2026/10/22/ai-agen…
The internet is genuinely melting down.
An urgent notice from the Railway team
blog.railway.com/p/security-…
1
12
6,377
ashwini asokan retweeted
okay guys that’s it, turn off the internet
🚨 Nightmare day for cybersecurity
Multiple major security incidents, active exploits, supply chain compromises...
If you run production systems, start patching NOW.
PHP: Two critical unauthenticated RCEs. Attackers already exploiting them. Patch NOW
LiteLLM: v1.107.0 on PyPI and Docker is BACKDOORED. Steals cloud, SSH, and AI credentials. Pin 1.106.2. Rotate secrets.
eMudhra: A certificate authority's keys were crackable. Update your browsers, systems, and trust stores.
Node.js / Next.js / React: Security releases for Brotli RCE, TLS bypasses, and TWO critical Next.js RCEs.
TiDB Cloud: PingCAP breached through stolen credentials. Customer information and other sensitive data exfiltrated. Read the attacker's release, fun stuff.
Plex2Shell: Weaponized within HOURS of the PoC. Attackers already deploying cryptominers and proxy backdoors.
BeyondTrust / Ivanti: Actively exploited zero-days added to CISA's KEV catalog. BeyondTrust attacks linked to a China-nexus campaign.
And that's not even the full list.
Patch. Audit. Rotate. NOW.
Best version yet, of whats going on in tech in ‘26!
On vibe revenue 🤯, consensus investing & how pvt markets are looking like public markets
1
10
2,131
ashwini asokan retweeted
It's very easy to feel productive with AI
There's always an output, an md file, an html file, a vercel link
You end up feeling like you've done a hard days' work but all you've done is use up tokens and disk space
ashwini asokan retweeted
by this time next year, most of labs revenue will come from what they build in-house and then sell further up the stack. the average users just don't
1. have deep enough pockets
2. work on hard enough tasks
gpt-synopsys and anthropic wet lab are the first of many products which come as a result of labs gatekeeping and utilizing superintelligent swarms to create new and capture existing markets. currently by use of footholds and in the future independently. by 2030 labs will have eaten practically every economically valuable sector such as finance, legal and healthcare.
what's to stop labs? why wouldn't they? how can a 10,000 employee healthcare company stand a chance if they are up against 10 million vastly superior agents to the swarms publicly available?
the labs understand this and are hence preparing for the "grand capture" by donating shares to the public (anthropic and spacexai already committed with openai about to) to minimize p(public revolution).
the question becomes what to do. one option is obviously to join a lab. if you want to join the lab I believe the best "unit chance of joining lab/time" ratio is proving scaleable user acquisitions and publishing relevant hot topic things on X.
if you don't want to join a lab, acquiring factories, propietary data, licenses, land and things that are human facing and exist in a stack of value creation will be fruitful as they become the primary bottlenecks once suffcient swarm cognition is solved
OpenAI, Anthropic and other leading AI labs face a growing challenge: Better models aren't necessarily attracting more usage. A 10-point intelligence advantage now earns just 20% more developer traffic, down from nearly 69% in late 2025.
Full story: thein.fo/4hGJ9jY
