How estimations (and fights around them) are changing when synthesizing a plausible solution is cheap?? @Neuralab was on both ends of the spectrum; both giving or auditing tons of software estimations, and it's funny how some things never change. You can thank Hofstadter's Law for this and @BZagor's essay explains why @ The Recursive 👇 therecursive.com/ai-can-make…

Yesterday I got a chance to speak on the first ever @InfobipShift Think Thank! @Infobip gathered 38 of us to exchange ideas and current state of affairs. So, if you’re worried about Jacob Coxon’s world destroying machines, or some couch-epidemiologist-turned-AI-influencer prophecies, here’s what the AI field and software truly looks from the inside: ALL teams are experimenting, tinkering, hacking on how to work and how not to work with LLMs. ALL think that current "metrics" are BS (new features shipped, new apps built, tokens burned, lines of code synthesized ... nobody cares). NO ONE is letting agents work on production code without supervision. Some developers write code, some read code, some build testing frameworks around the flows, but what’s clear is that formality and control in software production is here to stay. A connected takeaway is that finance, ops and legal teams are starting to place predictability guardrails on usage and model spending. This makes LLM implementation a highly specific, very niche and essentially non-transferable endeavor. I think this is all a repercussion of SPECKLED intelligence (what works for me might not work for you; and what works for me on some things, could prolly not work on some other tasks) We also talked about the negatives - a lot: The burnout is real. AI is introducing novel problems into the space. Devs now need to read 20 pagers from PMs and complexified PRs from peers. PMs need to read 20 pagers from stakeholders and HR is drowning in CVs as mini novels etc. Hence, the aversion towards AI on some tasks is also real. It’s now considered tacky or rude to send a prompted paragraph toward your colleague. Folks working on content inside teams are starting to forbid AI tools in writing. The crunch is also real, but I didn’t pick up that it is due to AI. Most of the crisis is connected to covid over-hiring and post-covid layoffs, non-ZIRP, inflation fears impacting budgets and AI sucking the air from every non-AI software room. In essence, the impact of AI on overall SWE productivity is still a big puzzle and I think it will be a puzzle for a long time. Which leads me to the 2nd picture here. Ema Puljak (@CERN) made the best, if not THE best, talk on Shift and the topic was Quantum Computing. What knocked me out of the boots was what Ms. Puljak placed on her first slide. She explained the difference between the HYPE and the REAL state of quantum computing, which was a refreshing framing. My prediction is that this talk will win the “talk of the shift award” - and I hear you will get a chance to watch it - but I will end this note with another prediction. The zeitgeist I picked up from Shift is that engineers and researchers are starting to talk about really specific and granular use cases, mostly ignoring/ridiculing the doomer and boomer narratives. So maybe the 2026 Shift was a tipping point where the AI convo turns to engineering, exactly where it should be.

I just got an email that @Neuralab has been approved into Anthropic’s Partner Program! We’ve been experimenting with models for years now, and have implemented LLM-empowered workflows both in and outside of the house. We know where LLMs shine, where they stutter, and we also know they have a killer instinct of a Tetris game. Clients, partners and makers need to know their limits and I’ll be boring everybody to death about it, starting tomorrow at @InfobipShift Think Thank. (Which is also the start of the physical book tour. ) Heftier news about the Claude partner program and book dates coming up sooooon 👋

Hey all, I'll be speaking at @InfobipShift conference on Monday (Think Tank Panels, from 2 to 4 PM). I hear that a lot of talks will probably want to place AI everywhere; but I think the Shift crew will appreciate at least one person asking where to leave it out 🙃

New stuff! We've started up a new collaboration with The Recursive and the plan is to publish hard-earned business truths about software projects. @Neuralab team mostly works on large-scale web and eCommerce projects so the first write up is about one frequent name in 2026 - B2B eComm. Couple of datapoints from my perspective: - B2B eCommerce is roughly half of our production nowadays, B2C is taking around 25% and various web portals or applications take the last quarter - Enterprise B2B is a tough nut to estimate. You're operating in thick fog of war, non-connectable ERPs or CRMs and access right policies that have more volumes than Alan Ford. - Most pressingly, and the main pillar for Bruno's Recursive column, stakeholders think they can just deploy a clone of their B2C eComm and slap a membership module on top of it. - as with all SHORTCUTS in software making, it does not work So, click on the piece, read why you need a team of experts, or comment if you think otherwise - therecursive.com/b2b-ecommer…