... implicitly tuned to a specific model version and don t generalise cleanly to the next one. When we upgraded from gpt-4o to gpt-5, we actually saw a drop in performance. Having proper evals helps :) Comment from Musa Molla(@musa

... lose sight of why we started If you re building, reviewing, or just trying to keep quality high while moving fast you re not alone. Keep shipping. You re already doing the hard part. Check it out : https://www.graphbit.ai/ Musa ...

... behind each suggestion. That turns review from a checklist into a conversation about the code. We built it so teams can move faster without losing understanding. If you re curious, here s the link : https://www.graphbit.ai/prflow Comment from Musa Molla(@musa

... Substack I'll link at the end, but to avoid UI hopping I wanted to provide the basic summary here. If you've been trying to build something but you're fighting against entrenched user behaviours with a competitor or feel it's just too hard to go up against a giant. Well, maybe you have a secret advantage. Keep reading... What's changing? The internet was built to be: Human-read, human-clicked, human-navigated. Everything, like everything ... ... this future, everything you are describing is true. However we are far from that fully realized future and the problem with the jagged edged development of agent capabilities, is that humans never know exactly when to trust an agent vs

... keep re-explaining the same things, especially when work passes between teammates. Many teams use Intrascope alongside coding tools. Intrascope handles the thinking, planning, and coordination layer, while dedicated coding tools handle implementation. Comment from Vladimir(@vkrstic): @musa_molla  Absolutely. We noticed the same shift very quickly. Once AI becomes part of daily work rather than an experiment, multiple tools, keys, and scattered contexts create coordination and cost visibility issues. At that stage, teams benefit far more from shared ... ... models. Comment from Vladimir(@vkrstic): @busmark_w_nika  Absolutely agree. That s exactly what we saw during the launch as well. Depth of conversations and real usage feedback mattered far more to us than raw numbers. Comment from Musa

... them, only the preview. That felt annoying to be honest Comment from Victor N(@viktorgems): @busmark_w_nika  I did but nothing changed, it was together with other 2 questions and this one was somehow ignored Comment from Musa Molla(@musa

... barrier to try a mobile app / product is higher for me. I'm a desktop primary kind of person, so if it's mobile it has to really resonate. I suspect this may track somewhat with age? Comment from Musa Molla(@musa

... deeply. I've been on both sides of repeated review comments and it drains energy fast. Automating the obvious checks lets me focus on real design and logic decisions. That's where reviews actually help me grow. Comment from Musa Molla(@musa

... This really resonates with me. I don't miss issues because I don't care, I miss them because my brain is overloaded. Anything that clears basic distractions lets me spend energy where real judgment is needed. Comment from Musa Molla(@musa

... monatruong_murror): @yellow_yetti  Happy new year Vera, love your message <3 Comment from Mona Truong(@monatruong_murror): @busmark_w_nika  I hope all your plan in 2026 will be achieved. Nika <3 Comment from Musa Molla(@musa

... reviews optimize for clarity. If future-me can't understand it in six months, reviewers will call it out. That taught me fast that readable code matters more than clever code on my team. Comment from Musa Molla(@musa

... best reviewers don t act like gatekeepers. They act like collaborators : asking questions, surfacing alternatives, and making intent explicit. The goal isn t approval. It s shared understanding. Curious to hear from this crowd : What behavior turns a review from friction into collaboration for you? Comment from Musa

... exclude other "less promising" ideas. If s omeone starts messaging you about your great idea , you can spot early interest, and when he/she wants to use it and pay for it, it is a win! Comment from Musa Molla(@musa

... reasoning stability over long contexts , not just single-prompt benchmarks. Happy to share concrete findings if helpful this is exactly the kind of model where early stress-testing pays off before v2. Comment from Chirag Arya(@chirag_a2207): @musa_molla  I completely agree with this perspective. Once a model becomes part of an execution path, its predictability and failure characteristics are much more important than its peak benchmark scores. These are the specific aspects we are focusing ... ... million tokens to investigate any of these system-level behaviours, we would greatly appreciate your feedback. Observations from real workflows help shape our priorities moving forward. I'm happy to share notes as you conduct your exploration. Comment from Musa

... coming year Comment from Boris Glen(@borisglen): @busmark_w_nika  Nika, that's right, I hope so :D Merry Christmas! Comment from Nika(@busmark_w_nika): Yeah, and eventually, leading to more MRR! :D Merry Xmas! Comment from Musa Molla(@musa

... Having my fingers crossed for you! :) Comment from Nika(@busmark_w_nika): @bekjon_ibragimov  you can reach out to someone, maybe someone who uses Goodreads or some Reddit forums about book fandom. Comment from Nika(@busmark_w_nika): @musa_molla  I have that problem stepping out. Usually I only take a look at details, so I am overwhelmed by how a few things I accomplished (because I lack the whole). Comment from Lea Yi(@lea_yi): Thanks ... ... emotions as a trigger to form a micro-saving habit. As for the co-founder part, it's still in the discussion stage. I'm not sure if I'm bringing that person on board for sure. Comment from Musa

... disappears. Shared vocabulary emerges. Work becomes more legible. AI isn t just changing execution. It s changing how teams talk. Curious to hear from this crowd : What term or concept became clearer once AI entered your workflow? Comment from Musa Molla(@musa

... commands become a negotiation between : intent, interpretation, context, and constraints. Interfaces aren t static , they re conversational. I would love hear from the crowd how often does instructing an AI feel like a negotiation in your workflow? Comment from Musa Molla(@musa

... executes the method. Teams own the consequences. This clarity is changing how teams think about responsibility. Curious to hear from this crowd: What part of accountability has become more important since Agentic AI entered your workflows? Comment from Musa Molla(@musa

... system can now do natively. The real productivity gain isn t automation. It s rethinking the architecture entirely. Curious to hear from this crowd: What s one workflow you rebuilt because AI made the old version irrelevant? Comment from Musa Molla(@musa

... reliability itself, but there are more memory frameworks emerging as of late, which I think will do great things for adoption and moving things forward where they've stagnated as of late. Comment from Hayds L(@hayden_lawson): @musa_molla  Bingo! Comment from Musa

... predicting your biggest AI failure? Comment from Abdul Rehman(@abod_rehman): Measuring only accuracy is a trap. I ve learned the hard way that tracking drift and anomaly patterns is where you actually see failures coming. Comment from Musa Molla(@musa

... breakers instead of blind retries It s not about making AI smarter, it s about making it airworthy . So here s the question: If AI systems had to pass a flight test before production, what would your checklist include? - Musa Try Graphbit here: github.com/InfinitiBit/graphbit Comment from Sanskar Yadav(@sanskarix): Love this analogy! So if AI had a flight test, my checklist would include: state persistence after every step, clear rollback points, automated diff checks for output consistency ...

... intelligence, it s autopilot. Now I m curious, What s the moment when your AI system felt stuck ? Still working, but not growing? Let s talk about the problems that don t throw errors but stall evolution. - Musa github.com/InfinitiBit/graphbit Comment from Nosheen Kanwal(@nosheen_kanwal): Richer feedback = Richer additions or deletions It doesn't have to stay same just because it stops giving errors ...

... making it trustworthy . Now I m curious: What s the sneakiest it still runs but gives the wrong answer bug you ve ever seen in AI systems? Let s share the stories we never put in the postmortem notes. - Musa Comment from Ghost Kitty(@ghost-kitty-7815048): Comment Deleted Comment from Musa