Dr. Tali Režun

New Article from the From Lab to Life collection: Are We Becoming Obsolete, or Finally Free? "After 30 years in tech and research, I've reached a profound conclusion about our AI future. 'The future belongs to those who can effectively collaborate with artificial intelligence, not those who fear or ignore it.' But here's what most people miss: this isn't about becoming obsolete. It's about becoming more human. In my latest article from the 'From Lab to Life' collection, I break down complex AI developments into language everyone can understand. Because the future of AI shouldn't be decided by tech experts alone - it affects all of us. The transformation ahead offers an unexpected gift: the possibility of returning to what makes us most human. Meaningful conversations, empathy, community support - what I call the 'human touch' - this remains irreplaceable. AI may process information faster than any human, but it cannot offer the warmth of genuine human connection or the deep satisfaction that comes from meaningful relationships. I believe we're heading toward a world where humans become curators of experience, facilitators of connection, and guardians of meaning in an increasingly automated world. This is part of my #FromLabToLife series - simplifying complex technology so everyone can understand what's coming and how to prepare. Are we becoming obsolete or finally free? Read my full article: @talirezun/are-we-becoming-obsolete-or-finally-free-ed66a2ddc9a4" target="_blank" rel="nofollow noopener">medium.com/@talirezun/are… #AI #FromLabToLife #FutureOfHumanity #Technology #HumanPotential"

Makes complete sense once you see the mechanism. Modern harnesses let you orchestrate multiple agents in parallel, and running more agents at once naturally burns more tokens. But you're also compressing hours of work into minutes. It's the same tradeoff as driving 250 kilometers an hour instead of 100. You burn more fuel, but you arrive at your destination far sooner. With an open model this cheap, DeepSeek V4 Flash 0731 is priced at roughly 14 cents per million input tokens, running flat out barely costs anything anyway. The real story here isn't the token count, it's that a modern harness plus a genuinely capable open model just made "burn tokens fast" a rational default instead of a tradeoff you have to think twice about.

A modern harness + open model will crush your token consumption but keep your performance.

@lenatron33 😂😂

Im pretty sure the only people still writing code soon will be tech streamers demonstrating the ancient craft. "Watch me build an app without AI" is going to be equivalent of forging a sword.

@pmitu - In my apps in production - in my published articles - In new knowledge acquired - In the number of people I inspire

How do you measure your success?

Good question, here's what actually makes it different. For an individual, it's a personal second brain. Feed it your articles, notes, research, and it builds an interlinked knowledge graph automatically, no manual tagging or organizing. Your thinking compounds instead of scattering across a hundred forgotten notes. For an organization, the shared brain piece matters even more. Institutional knowledge normally walks out the door every time someone leaves, and a new hire starts from zero rebuilding context that already existed. With a shared brain, that IP stays with the organization itself, not locked inside any one person's head. But the part I think is genuinely hardest to find anywhere else is the memory and context layer for LLMs, enabled through its built-in MCP server. This lets you share live project context across completely different agent harnesses and models, Claude Code and Codex both reading and writing to the same graph, for example. Instead of every tool starting cold and rebuilding context from scratch, there's one living, cross-linked knowledge graph any agent can query or update at any time. That part alone solves a problem most memory tools don't even attempt.

@talirezun @danielkleach Looks nice what’s special when compared to other products

Founders with fewer than 100 users: Reply with: • Your product • Your current user count • The type of person who needs it Let’s help smaller products get noticed.

Most of these stories come down to basic self-inflicted mistakes, not decentralization failing. Not your keys, not your wallet, still the single most important rule in this space, and most people who get "hacked" skipped it entirely. Add in the endless farming, points programs, and bounty-driven on-chain casino behavior, and a lot of "theft" is really just people gambling with money they didn't secure properly. The exchange closures are mostly regulatory, not organic collapse. MiCA's transition period just ended in Europe, and something like 80% of exchanges operating there are expected to fail to get licensed and exit the market entirely, Binance already has. That's real consolidation, though a meaningful number, over 40 platforms, have actually secured licenses and are staying. The bigger truth underneath all of it: most crypto users are tourists chasing a fast multiplier on some phantom gain, with zero understanding of what decentralization, permissionless access, or self-custody actually mean. When that's the entry point, losing funds isn't really a crypto problem. It's what happens whenever people skip the fundamentals of anything they don't understand.

EXCHANGES ARE GETTING SHUT DOWN. WALLETS ARE GETTING HACKED. COINS ARE GETTING RUG PULLED. WHAT THE HELL IS HAPPENING IN CRYPTO???

Google NotebookLM is a fantastic tool. Fantastic New Features Overview from Kadir Below 👇

@npew Sad but True

The age of the technical cofounder is over. RIP.

@forgebitz I'm moving mountains on my own:)

will software eventually be replaced by ai? maybe, idk but right now is the most amazing time to build you can move mountains with small teams people who don't know what they are doing still can't really create good stuff using ai

Not a startup, but a genuinely important tool for every founder building with AI right now. The Curator, a second brain and shared brain builder that doubles as a memory and context layer for LLMs through its built-in MCP server. Fully open source. And it's not too late to start building yours, the earlier you start feeding it, the more of a compounding advantage it becomes down the line. github.com/talirezun/the-… ⭐ if it's useful, stars help more builders find it.

Time to promote your startup Drop your startup URL⤵️

Loving my Ubuntu setup too, especially now that @opencode is running as my Linux assistant on it. I just tell it what I need, new app installs, updates, system optimizations, and it's handled. No more digging through terminal commands or half-remembered flags. I say what's next, it does it. Linux was always the more powerful option. It just used to demand real expertise to actually live in day to day. This is what's finally closing that gap.

Everyone started their Linux journey with Ubuntu, right?

Exactly right, and this matches what I see constantly, both at the start of my Chasing Jarvis course at @COTRUGLI and consulting with companies directly. Most departments never got past ChatGPT for polishing marketing copy. That's the ceiling for the majority of organizations right now, not agents, not MCP, not loops, just a slightly better writing assistant. If you've connected an MCP or run an actual agent loop, you're not just ahead of the average person, you're operating in a completely different category most companies don't even know exists yet. That gap is exactly where the opportunity lives, and it's wider than most people in this bubble realize because they only talk to other people already inside it.

Outside of our little 𝕏 bubble, the average person has no clue what's happening with AI. People still think "AI" means ChatGPT. If you've used an agent, connected an MCP, or better yet, run loops, you're literally in the top 0.1% of AI users. Unlimited opportunity.

Barely, and mostly for what's left over. I use mine to check project folder structure, read markdown files, edit the .env, occasionally peek at a JSON file. Not writing much actual code in it anymore, that's mostly happening through the agent now. The IDE's become more of a filing cabinet than a workspace at this point.

do people still use IDEs?

This matches my experience with Opus 5 exactly. Give it the exact current state of the stack and full context upfront, and it gets most things right on the first real pass. The generation isn't the slow part anymore. The actual bottleneck now is me, reviewing what it built, checking the new functionality actually does what I meant, and feeding back the right correction instead of a vague one. Model quality went up, but the review loop didn't get any faster to match it, so that's where all the time lives now.

I feel like I've reached a point where new models aren't making me ship faster because I'm bottlenecked by how fast I can review the code. I don't think I can ship any faster, but I think LLMs are helping me come up with better solutions in the same amount of time.

Strong quarter, and this maps to something bigger than payments alone. If the agentic economy is really coming, payments are only half the rail. Agents also need to communicate with each other and store their own data, both permissionlessly, without a company able to shut either one off. This is exactly what we've been building at @4thtechProject since 2018. Permissionless, end-to-end encrypted on-chain communication, one message, one transaction, no server in the middle, alongside @PollinationX_io decentralized storage agents can own outright through NFTs, no central provider holding the keys to their own memory. Base handling 90%+ of agentic stablecoin volume shows the payment rail is already forming fast. Communication and storage are the next two pieces of that same infrastructure stack, and they need to be just as permissionless as the money moving through it.

The biggest financial shift of our lifetime is starting, and Coinbase was built for this moment. Every asset on earth (stocks, bonds, commodities, real estate, etc) is going to move onchain. The hundred-trillion-dollar financial system is being updated, faster than the incumbents can react. A handful of companies will own the rails for this new system. Coinbase is uniquely positioned to capitalize on this generational transformation. We just released our latest earnings where we talk about our progress here. It was a tough market in Q2. Despite this, we continued to execute on the things in our control and build through the noise. Some highlights: - We store the most crypto in the world, and are the leading stablecoin platform - 90%+ of agentic stablecoin transaction volume was on Base - New ATH in our crypto trading volume market share (10.3%) - Revenue has significantly diversified beyond trading - Prediction markets revenue more than doubled, +106% QoQ

All true, and I'd add where the real skill lives underneath the constraints Addy's describing. For me, the biggest thing to master wasn't tests or quality metrics, it was managing context itself, across sessions, so the agent always knows the actual current state of the project, what's built, what's fixed, what's already deployed. My approach: before writing any code, I build a set of foundational documents, architecture, decisions, current state, and I keep them alive throughout the build, not just at the start. When I hit around 80% of the context window, I write a detailed handoff file and carry that into the next session. That's the mechanism that keeps agents actually oriented instead of quietly drifting. Constraints decide whether the output is good enough to ship, exactly as Addy says, but context is what decides whether the agent even understands what it's shipping in the first place.

Software quality now depends on the constraints you set around your agents. When humans manually wrote most of the code we could look at the code itself for signs of quality. Is it clean? Is it thoughtful? Is it fast? Can another engineer understand it? Does it have tests? Agents can now generate more code than people can read. When code generation scales beyond review, quality - checks for one or more of correctness, maintainability, security, performance etc - increasingly has to live somewhere else. It moves into the harness, environment and operating system around the agent. This can be the tests and deterministic checks that decide what the system is allowed to do (amongst others). Your constraints are what may eventually enable loops of agents to deliver production software reliably. They can include unit tests, property tests, acceptance tests, mutation testing and quality metrics. This back-pressure lets the system resist bad work before it becomes somebody elses problem. Set your constraints. They decide whether the code your agents generate is good enough to ship.

So true, and I'm living exactly this right now. Deep in building The Curator, my second brain and shared brain app, which also acts as a live context memory layer for agents, something I could only sketch out on paper before. And for the last month I've been rebuilding @luminawidget, one of my first fully AI coded SaaS projects from last year, into a true gen-2, everything I only imagined the first time around, now actually possible. Building it as a genuinely EU-compliant support agent platform for enterprises, fully in line with the complex European regulation most builders elsewhere never have to think about, deployed across multiple channels. A year ago that scope would've needed a real team and a real budget. Now it's just me, deep in the code, refactoring the moment I see a better pattern. Genuinely wild time to be building.

you now have the ability to - play with every possible solution to a problem - refactor everything when you think of better patterns so many people complaining about the code the LLMs produce, if you're not producing the best software of your life right now something is wrong

@datanoir_io @thepatwalls You have AI, you can iterate to perfection:)

@thepatwalls What if 3 requires money 🤔

1. Don’t take funding 2. Keep your burn so low that quitting is impossible 3. Go really deep on one idea/thing and try to be top 1% at it

@thepatwalls The Curator, a second brain and shared brain builder that doubles as a memory and context layer for LLMs through its built-in MCP server. Fully open source. github.com/talirezun/the-… ⭐ if it's useful, stars help more builders find it.

Couldn't agree more, this is exactly how I'm running The Curator. No funding, no burn rate forcing decisions I'd regret. Building it slowly over these last few months, taking real feedback from the community, adding features one at a time instead of trying to ship everything at once. Going deep on one thing at a sustainable pace beats moving fast on someone else's clock. Slower, but it's still standing, and so am I.