Get live statistics and analysis of Lamb's profile on X / Twitter

once you understand how gpt works with context windows, token probabilities, and hardcoded system prompts, you stop mistaking prompt theatre for engineering prompt theatre (overcomplicated scaffolding) often looks impressive but doesn’t improve model performance in meaningful ways. gpt is surprisingly robust to simple, direct instructions most “prompt engineering” posts are performative and engagement farming bloat and jargon in prompts don’t make gpt smarter they just eat context, dilute signal, and increase token cost long prompts reduce usable context, and unclear instructions can confuse the model or trigger irrelevant associations the model isn’t “thinking” it’s just matching patterns you're better off going through @apollonator3000's twitter and searching for prompts that fit your need or get a primer on context profiles

been deep diving into likely trends that will come up in 2026. here’s what’s coming fast and underpriced. lock in: 1. mcp-native apps: apps that speak model context protocol out of the box. zero glue code. zero wrappers. agents just work. see langchain, and also what the team at pipedream are doing with string dot com 2. rag-powered doc agents: agents that don’t just read documents, they interpret, summarise, and act on them. contracts, invoices, playbooks = parsed, tagged, and executed. huge if you're in the healthcare/legal/finance subniches docugami is cool but they have no taste, and ui is ass. inject some taste in this and you'll be printing $$$ 3. async-first workflows: systems that progress without meetings. agents handle handoffs, decisions, context. founders sleep, ops continue @boringmarketer and @gregisenberg been going heavy on this idea 4. edge-native ai: ai that runs in-browser or on-device. no cloud calls. no vendor lock-in. faster, private, local-first automation 5. hyper-personalised automations: agents that adapt to user behaviour in real-time. one user, one ops system. personalisation that actually learns. @apollonator3000 has already touched a bit on this with his context profile lessons. this will soon become a moat imo. reclaim ai is already testing this with smart calendar + task scheduling that adapts to your work patterns 6. multi-agent biz ops: teams of agents handling cross-functional workflows. sales, support, onboarding, finance. not one bot but ten specialists. think factory ai but instead of swe, think for biz ops 7. real-time automation triggers: agents that monitor live data streams and act instantly. no cron jobs. no polling. just event → decision → output ppl shit on n8n but given the direction they are going in, i can only see their business getting bigger 8. browser-based ai clients: chrome is now your runtime. ai tools live in tabs. no install. no infra. just extensions doing work in the background with how great google has been doing with gemini and how fast they've been shipping SoTA products, this will soon be a reality 9. api-first infra: every function, every agent, every tool = an api. composable, swappable, pluggable systems. infrastructure as lego blocks few 10. no-moat saas clones: apps with basic crud and no unique logic get replaced by agents. if chatgpt can rebuild it, it’s already obsolete. 11. ai-native onboarding flows: not walkthroughs. conversations. agents onboard users by asking questions, configuring tools, and routing data live. 12. self-optimising agents: agents that refactor their own workflows over time. usage → feedback → optimisation without developer input. 13. private ai stacks: local-first agent deployments. no data leaves the browser. critical for regulated sectors and control-obsessed founders been advising solofounders and local sme's for this. pays well 14. agents that manage agents: meta agents that delegate to sub-agents. coordination layers, not just execution layers. ai as operations manager. think Claude Code subagents. major cheat code. 15. ai summarising workstreams: agents embedded in tools that generate live, contextual summaries: what changed, what matters, what to do next. cursor are moving in this direction in order to compete with anthropic, and it's looking good ngl 16. chrome becoming an os: extensions + edge models + persistent agents = chrome becomes the new desktop. your workspace is now browser-native. google won't let perplexity win this agentic browser race, they'll lock in after gemini 3 arrives 17. autonomous backoffice: from invoice gen to pipeline sync to report prep. fully automated operations run by agents, not staff. 18. open-source agent swarms: agent networks you can fork, remix, and deploy. not just tools, but ecosystems. npm for intelligence seeing some noise already on github and hugginface 19. motion-ui for ops ux: invisible agents need visible feedback. animated ui gives trust, state, flow. motion = clarity when logic is hidden if you have taste, this and good aesthetic, lock im 20. rag + mcp = new web standard: retrieval for memory. mcp for action. together they replace the backend. this becomes the new base layer of the web. i’m not only building around these, i’m building with these ideas in mind

Fear often holds us back from this ONE thing we've desired our entire lives. An obstacle which every human must endure. But imagine you had no fear... What kind of human would you be? Homage: @DejaRu22

most ai saas businesses are underwhelming because they try to create alternatives to the human process what do i mean by this? they aim to build tools that perform tasks a human would find far too complex to do think: "an ai sdr that can cold call 8 of your prospects at the same time" or "ai that can create unlimited ad creatives in minutes" most of these tools, while technically impressive, enter the market with velocity because of the hype cycle we're in. they sound novel, and because we've been pre conditioned by big players to expect state of the art products, we try them with high expectations and end up dropping them after a few weeks look at flowith for instance. their agent neo was supposed to be the greatest thing since sliced bread. it built major hype and now we hardly hear about the product the best ai saas are the ones that map out the human process behind the magic cursor, manus, perplexity, etc all managed to stick around because of the human heuristics that govern their process you watch them perform tasks you can imagine yourself doing, but at speed, which makes you feel the agi like @karpathy said, the best agents are akin to "human spirits" and i completely get that. they stick with you because they feel like genuine assistants you can map out their entire process, how they go from a to b to c to d, not from a to z without any clear mapping the best agents are built with human characteristics in mind. they make you feel part of the process, so after a while, you feel safe to let them run and do their thing the best n8n templates i’ve seen nailed this process if you’re building an ai saas, build with this in mind and you’ll have lower churn rates