farez πŸ‡΅πŸ‡Έ (@farez) header
farez πŸ‡΅πŸ‡Έ (@farez) profile picture

Building software and AI for UK gov. Also building: - https://teapawt.net/t.co/5cEhewTuv8 - PDF/DOC to MD/HTML (launching soon) - https://teapawt.net/t.co/DMCZ6bxCIQ - invoicing for Notion

London, UK
Joined May 2007
Photo by farez πŸ‡΅πŸ‡Έ (@farez)Photo by farez πŸ‡΅πŸ‡Έ (@farez)Photo by farez πŸ‡΅πŸ‡Έ (@farez)Photo by farez πŸ‡΅πŸ‡Έ (@farez)Photo by farez πŸ‡΅πŸ‡Έ (@farez)Photo by farez πŸ‡΅πŸ‡Έ (@farez)Photo by farez πŸ‡΅πŸ‡Έ (@farez)Photo by farez πŸ‡΅πŸ‡Έ (@farez)Photo by farez πŸ‡΅πŸ‡Έ (@farez)Photo by farez πŸ‡΅πŸ‡Έ (@farez)
I like where Alex is going with his new service business. Like a lot of tech founders, we believe there's a definite expiry date now for SaaS businesses in its current form. Nobody really knows where we're heading. But what is still certain in the near future and now is the need for experts to help specific parts of businesses excel. And this means consulting. And not just consulting. I'm talking consulting + agents. Businesses want to save time and increase profit. This usually means automating away processes that take up human time. And building automation require: a) tools to automate b) the technical skills to build the solution using those toold c) the consulting savvy to understand the client's process and present a solution that the client can understand and use easily This is not new. Consultants have been doing this for decades with software. Now they will do it again with AI agents. I don't know how long the demand for consulting + agents will last but for now, I believe it's the least risky and most financially rewarding route to $$$, if you're a tech founder. "But it's not scalable". Correct, it's not. But the alternative is working long hours on a SaaS or wrapper or app in a competitive category that can be eaten up by ChatGPT/Claude/Gemini in a blink. And not seeing a penny of revenue. With consulting + agents, we have a window within which we can provide a service, get well paid for it, and use the money to invest in other things with lower risk. I've been a software engineer and consultant all my life. I felt the pull to being a SaaS founder (who didn't!). But I believe it's a good time to be a consultant again. Or at least a least riskier. And I have been focusing on this for the last 6 months. Building a portfolio of AI consulting projects. It's a great time for service businesses.
Building a $1M for the second time. Day 1 of running my new 10X.ing service ✨ The beauty of running a service business is that you are dealing with people. Just sent personalized presents to everyone.
3
1
95
RIP Coding
It's an odd feeling. Two things I've spent the most combined time doing for the past 10 years: - Coding - Editing I've spent thousands of hours doing these - and they're easily the two things that have been the most defining for the success I've had. And they are now two things I'll probably never do again. Haven't written a line of code for almost a year. Haven't touched an editing software for +2 months. And I think I never will again. They were 10 amazing years though. I'll never forget them.
61
Just added @PopInvoice to ChatGPT Business. Looks like it could be useful for traffic analytics.
53
ngl... I used to want to build startups (so I can flex my engineering muscles and make useful things) These days I'd want to be employed again. By frontier AI companies. Because that's where all the cutting edge stuff is happeing.
2
118
farez πŸ‡΅πŸ‡Έ retweeted
Claude coding tip: Use /loop If you have subagents running on tasks that will take hours then get your main agent to a check-in every 5 minutes. Why? because sometimes subagents stall or get stuck. Example: /loop 5min Check-in on all helpers, check that they're progressing, close helpers and clean up their worktrees once they're done, start new helpers to work on the next tasks if needed. Achieve our goal <X>.
2
1
1
53
TL;DR AI is better than humans at making another AI smarter.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧡 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
80
farez πŸ‡΅πŸ‡Έ retweeted
I'd join a low MRR club. It would feel like going to a low or no alcohol bar - conversations would make sense and I won't be hating myself in the morning.
1
1
1
44
Some of our users (of popinvoice.com) are experienced software developers. Not everyone wants to vibe code and maintain their own SaaS.
3
2
914
API key house cleaning
1
631
I'm quite sure I'm in a bubble here, but not building with or for AI at this time in our lives would mean missing the biggest opportunity of our generation to create something amazing and make a big difference to humanity. If I was retired, I'd come out with time and build. If I was already successful, I would still get my hands dirty and build. If I couldn't afford a computer, I would borrow one and build.
409
Build: ❌ Faster βœ… Bigger βœ… Different
1
3
289
Is a "tool that Claude can use" the new SaaS?
If Descript completely axed their built in agent "Underlord", and instead exposed the tools and let Claude Code and Codex edit in their platform directly it would be the fastest growing editing platform in the world. Their transcript tech is world class, but they force you to use their built in agent. They will get cooked for this. Can someone please build the agent native descript. Please i will invest.
2
288
Want to know what your agents can see? I'm building something that will tell you. Sharing it here soon.
2
189
Yep, Andrej's suggestion works well. A family member is needing tracheostomy (a tube into the throat to help with breathing) and I was researching how this impacts speech. I got Claude Opus to turn my findings into an explainer video I can share with my family. Asked it to use my Elevenlabs account to generate the narration. This took 10 min (including integrating Elevenlabs into Claude Desktop). Can be improved but what a great way to share and explain.
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work! In summary: - As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. - Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
1
197
Why are so many programmers/developers so unhealthy? They usually look like they'd run out of breath just climbing up a flight of stairs.
147
Experimenting with building a Claude mod that tells watches how you work and determines if you need a break :)
You can now mod Claude Code: - Change how it behaves - Customize the UI - Swap in your own features Write one with a few lines of TypeScript, or have Claude build it for you. Mods ship inside plugins, so you install them with /plugin in the CLI or desktop app. A few examples:
176
How's your mood? I've noticed that my mood really affects a lot of the ways I think, make decisions, or analyse a situation. If I'm in a good mood, my decisions are always based on hope, positivity and opportunity. When I'm in a bad mood, my decisions are always based on fear or conflict avoidance, and an unwillingness to compromise. This is one of the reasons I avoid reading the news and drastically minimise social media consumption. These channels are great at getting me down, especially if I start the day consuming them. So now, whenever I need to make a decision on something or analyse situation: I check my mood. If I'm not feeling great or feeling negative, then I defer all that until I feel good again. That will usually lead me to a better outcome.
123
I've shaved my head and now selling ad space on it. Prices Sides: One slot on the left and one slot on the right are free. First come first served. Middle back: Auction starting at $10. Middle front: Auction starting at $5. Others. Auction starting at $1.
πŸ€– Made with AI
2
1
153