
Work: AI and what comes next #London Spare time: private AI dinners for VCs, founders, leaders & academics. + I like art.
London
Joined January 2024
- Tweets1.6K
- Following75
- Followers105
- Likes501
Pinned Tweet
AI coding tools are currently causing quite a stir among developers. In my opinion, Claude Code excels at reading entire codebases editing files effortlessly and even creating user interface prototypes using natural language. Codex, on the other hand, is better suited for autonomous refactoring and handling multiple file changes. It also boasts higher token limits and greater speed. Some developers prioritise creativity and intent understanding while others focus on speed and quotas. Ultimately, many developers blend these approaches for brainstorming and then execution, with benchmarks remaining close.
50
398
Many people are terrified of generative AI but history suggests we should be cautious about equating fear with foresight. Simply put, generative AI refers to systems capable of creating content like text images music video and even code. Rather than simply retrieving information these systems generate new outputs based on learned patterns. This is why tools like ChatGPT can write, Midjourney can create images and coding models can build software.
The panic around it feels familiar. People feared the printing press would destroy knowledge. They feared photography would kill painting. They feared calculators would destroy mathematics. They feared the internet would make society dumber. Every major technological shift arrived with predictions of catastrophe and what usually happens is less dramatic.
Technology transforms how we work, creating new opportunities while also eliminating some old ones. It becomes so ingrained in our lives that future generations may struggle to imagine life without it.
However, this doesn’t mean AI is entirely risk-free. Every powerful technology presents challenges. But viewing generative AI as uniquely dangerous overlooks a fundamental truth: humanity’s greatest advances have often seemed frightening to those witnessing them for the first time.
20
If the rumors are true, tomorrow could mark the beginning of a new chapter for Anthropic.
Claude Fable 5 is reportedly built on the same architecture as Mythos, the restricted Project Glasswing model that turned heads with 77.8% on SWE-bench Pro and 83% in cybersecurity benchmarks, outperforming models like Opus 4.7 and GPT-4. The challenge was never intelligence. It was making that intelligence safe enough for broad public release.
What’s interesting is how quickly the conversation has shifted. Not long ago, AI labs were competing on whether they could build capable models. Now they’re competing on who can successfully productize their most powerful systems without breaking trust, safety, or usability.
If Alex Heath’s reporting is accurate and Fable 5 arrives tomorrow, the benchmark numbers will grab headlines.
150
So @bottinogames posted a rant that anyone who wants or uses AI to write game code should “destroy every keyboard they own.”
I replied with a moderate take: nobody serious is asking AI to build an entire game for them. Most developers use it to speed up repetitive tasks so they can focus on the difficult parts.
Apparently that was too much🤣 but Not too much to debate… it was too much to tolerate.
What did they do? They blocked me.
Which imo is fascinating because if AI is really as useless and disastrous as some people claim, you’d think they’d be eager to dismantle opposing arguments in public. Instead, some seem to treat disagreement like a vampire treats sunlight.
The funniest part is that the people constantly shouting “learn to code” are now panicking because more people are learning to build things. Imagine being so confident in your position that a single polite reply sends you reaching for the block button.
1
22
Angelo Valentino retweeted
Replying to @NoemiTitarenco
The funniest outcome of AI is that people are rediscovering how formulaic human writing has always been.
Turns out humans also use patterns, clichés, rhetorical structures, and stylistic tics.
The existence of a pattern isn't evidence of AI.
Sometimes it's just evidence that someone knows how to write.
The growing tendency to blame every layoff solely on AI is becoming intellectual laziness. Companies were already overhiring long before modern AI emerged. Cheap money, pandemic demand, inflated growth forecasts, bloated organisational structures and years of poor management decisions created a workforce bubble that was bound to eventually correct itself. Now AI arrives and suddenly everyone wants a single scapegoat for every redundancy notice. Reality is simply less convenient.
Some jobs are being affected by AI. Some aren’t. Some companies are cutting costs. Some are restructuring. Some are fixing mistakes they made years ago. Some executives are using AI as a convenient excuse because “the robots took your job” sounds better than “we hired too many people.”😂😂😂😂
People enjoy simple explanations because they require less mental effort. However, unfortunately for them, the economy isn’t a Marvel movie and isn’t always straightforward. There isn’t always a single villain.
6
75
Many AI critics have quietly shifted their focus from debating the technology itself to apocalypse fan fiction. Every technological advancement is claimed to be a bubble, every investment a fraud and every company a scam. Despite being wrong for years they remain utterly convinced collapse is imminent.
The funny part is that actual frauds usually struggle to produce useful products😂, AI is doing the opposite.
Businesses are deploying it. Developers are building on it. Researchers are accelerating discovery with it. Governments are investing in it. Consumers are using it daily. Entire industries are restructuring around it. That’s a strange definition of a “rug pull.” What many critics cannot seem to process is that transformative technologies look expensive before they look obvious.
The internet seemed overhyped, smartphones unnecessary and cloud computing wasteful. Now the same people are pointing fingers at AI infrastructure spending, accusing it of fraud simply because they can’t differentiate between speculation and investment.
Please, not every large bet is a scam.
13
93
Before AI existed, companies were already spending millions producing forgettable adverts nobody remembered five minutes later. Bad storytelling did not arrive with AI. AI simply exposed how many people never had taste to begin with.🤣
The funny part is that critics keep blaming the tool for the creator. Give a talentless marketer a camera and they’ll make garbage. Give them Premiere Pro and they’ll make garbage. Give them AI and they’ll still make garbage. The common denominator isn’t the technology.
It’s them.
And if we’re being honest, people rarely complain when AI generated content is so good they can’t tell it was AI generated in the first place. Suddenly the issue isn’t “soullessness.” The issue is that the audience enjoyed it before learning how it was made.
1
23
I think that’s probably right, Not because ASI is guaranteed tomorrow, but because every major incentive points in the same direction. Governments want it. Investors want it. Companies want it. Researchers want it. The UK alone is pouring billions into AI while racing to become a global AI leader.
Artificial Superintelligence simply means systems that surpass humans across most cognitive tasks, not just coding, writing, or research, but science, engineering, medicine, and discovery itself.
For the UK, the upside is enormous: faster medical breakthroughs, more productive businesses, better education, scientific leadership, and economic growth measured not in percentages but in step changes. In the short term, it will make individuals more capable. In the long term, it could make entire nations more capable.
The people waiting for ASI to disappear are preparing for a world that probably isn't coming.
This quoted post is unavailable.
10
143
I like to take pictures of people, especially people who work in AI or technology out of work. In case we forget “It’s still about the people”
1
21
133
I personally find Saylor’s explanation a little too convenient. So because Bitcoin drops 22%, spot ETFs see $4 billion in outflows, and suddenly the culprit is AI? Anthropic raises capital. OpenAI prepares for massive fundraising. SpaceX pulls in $75 billion. Therefore Bitcoin falls?
That feels less like analysis and more like narrative construction after the fact, tbh…
Capital does not move through markets with the simplicity of a garden hose. Investors are not collectively selling Bitcoin to buy GPUs. Markets reprice assets for dozens of reasons simultaneously: risk appetite, macro conditions, liquidity, sentiment, leverage, and expectations. Blaming AI infrastructure for Bitcoin’s decline risks turning a complex market into a children’s story, and the irony is that if Bitcoin is truly the superior store of value its supporters claim, then a few months of AI excitement should be irrelevant. Real hedge does not need an excuse every time the price falls.
The most underrated shift in tech right now is the fact that millions of people are discovering they no longer need permission to build.
few years back, turning an idea into software meant learning programming, finding developers, raising money, or spending months building. Now people are describing products in plain English and shipping working prototypes over a weekend. Mutual fund dashboards. Crypto tools. SaaS products. Internal business software. All built through vibe coding with tools like Google AI Studio and Gemini.
Predictably, critics love to joke about “vibe debugging.” Lol. When the cost of creation collapses, innovation stops being reserved for programmers.
1
9
112
For years, people told everyone to “learn AI,” “experiment yourself,” and “build your own opinions.” Then the moment ordinary users start benchmarking models, running local setups, comparing outputs, and questioning popular narratives, some of the loudest voices suddenly become uncomfortable.
lol, authority is being challenged now. The UK AI ecosystem is becoming a fascinating example of this. More founders, developers, and businesses are testing models directly, measuring ROI themselves, and choosing tools based on results rather than brand reputation. The gatekeepers used to control the conversation. Now thousands of people can verify claims with a laptop and an afternoon.
The lesson is the same as it was in crypto:
Don’t trust the loudest voice in the room.Test it yourself… else a rug pull is sure.
The future belongs to people who verify, not people who repeat.
A quiet rebellion is starting to form in AI.
For the last two years, the assumption was simple: the future belongs to whoever spends the most on cloud compute. Now developers are asking a different question:
What if we do not need to?
Tools like MiniMax M3 are showing that cheaper models can sometimes match or even outperform premium options like Claude Opus on real coding tasks. In Kilo’s testing, M3 finished faster, used fewer resources, and during its promotional pricing was roughly 16x cheaper on input tokens. At the same time, developers are increasingly running local models through Ollama on machines as small as a $599 Mac Mini M4, giving them effectively unlimited usage for debugging, coding, and routine development work.
The uncomfortable reality for AI labs is that intelligence is getting cheaper faster than many business models expected. When companies like Uber are already capping AI spending and developers can save $5,000+ per year by moving work locally, in the end, the most efficient one supersedes!
Blimey, reading this job debate on X leaves me shaking my head. RAW argues that the freedom of an online business earning $100,000 a year is more valuable than a well paid nine to five office job offering $250,000, flexible holidays and no boss to answer to. Jason, however, insists entrepreneurship involves a round the clock grind and every imaginable risk while a proper job provides security. Replies mention missing the short hours after work or praising office perks, and a recent study confirms remote work increases isolation and strains mental health. Back in my day, a man would find one steady position and stick with it. Lol!
Healthcare is using AI for diagnostics and personalised treatment. Finance is deploying it for fraud detection and customer service. Manufacturers are running predictive maintenance and digital twins. Governments are experimenting with AI tutors, public service reform, and productivity initiatives aimed at lifting economic growth. Wow!!
34
191
Companies are discovering that the biggest value often comes not from one giant model, but from specialised AI systems working together. Vertical AI, multi model strategies, and domain specific tools are proving more practical, affordable, and reliable than chasing the most powerful model on the leaderboard.
18
171
For years, people treated AI as a futuristic experiment. Now the UK is discovering what happens when a technology leaves the lab and enters the economy.
Thousands of AI companies are being created, investment is pouring in, SMEs are adopting tools at speed, and forecasts suggest AI could contribute hundreds of billions to UK GDP by 2035. The conversation is also shifting from job replacement to workforce transformation, with demand rising for prompt engineers, AI generalists, and people who know how to work alongside intelligent systems.
And as it stands, businesses do not care whether a tool is revolutionary. They care whether it saves time, cuts costs, increases revenue, or creates leverage. That is why AI is spreading far beyond London and into every corner of the economy.
1
65
Damn, why is everyone so obsessed with modelsssss!! I mean, the obsession on Smarter models. Faster models. Bigger models! Whyyyyy??
one of the most interesting signals coming out of Y Combinator’s P26 batch is that founders are increasingly building everything around the models instead. Not too long ago, Forbes highlighted startups tackling the unglamorous problems that determine whether AI actually works in the real world: memory, identity, observability, compliance, insurance, and compute. Companies like Project are rethinking AI native computing architectures. Memory Store is building shared memory between teams and AI systems. Apollo Atomics is chasing compact nuclear reactors and already claims 20 gigawatts in commitments to support future compute demand.
This is what technological revolutions look like when they start maturing! The gold rush begins with people chasing gold.
Think about it.
1
2
63
