Real Alpha
Watch modeSubscribed
All questions

Question

Are AI agents ready for real work?

Of the 27 moments we cut that answer this, 25 say they’re ready for some work and 2 say they’re not ready yet. Sundar Pichai said on Sources (Oct 21, 2026) that agents can work across large codebases, while larger productivity gains still lie ahead. Diogo Almeida said on The MAD Podcast (Oct 15, 2026) that polished, preference-tuned responses can make people overestimate reliability, while errors still limit automation.

Where they stand

Ready for some work25

  1. The MAD Podcast · · 2 min

    Mistral says its models speed senior researchers up two to four times

    “it accelerates our most senior researcher by a factor two, three or four with our models.”
    Pierre StockVP of Science, Mistral AI1:14:53
  2. All-In · · 3 min

    AI-written code can outperform when agents run evals

    “We use agents and coding agents now for probably half the code that is written inside our place, and it is superior to what was written before.”
    Arvind KrishnaChairman & CEO, IBM20:08
  3. Sources · · 4 min

    Pichai expects a productivity leap from AI agents in 2027

    “I think if in some ways I view 2027 as a golden age of that productivity frontier for us”
    Sundar PichaiCEO, Google and Alphabet9:59
  4. Big Technology Podcast · · 6 min

    Recursive self-improvement still involves human researchers

    “you almost have agents helping AI researchers in some automated way work together to build these models.”
    Sundar PichaiCEO, Google9:51
  5. MTS · · 4 min

    Enterprise AGI is changing the challenge from capability to control

    Aceituno says agents reached human-quality reliability on five enterprise tasks in early 2026. The bottleneck has shifted from prompting to governing access and actions.

    Bernardo AceitunoPresident, Stack AI

  6. a16z Podcast · · 7 min

    AI cut a complex data-lake project from six months to ten days

    “And that took maybe about 10 days. In previous days, that would have taken maybe about six months for a single person.”
    Jad NaousCEO, Grepr24:04
  7. a16z Podcast · · 3 min

    AI query optimization needs benchmarks, not guesswork

    “it was able to ground that in performance benchmarks”
    Jad NaousCEO, Grepr29:03
  8. MTS · · 6 min

    Agents automate code but still face business decisions

    “It will at least confront the user with these high-level decisions because it still has business impact and outcomes to it.”
    Han WangCo-founder, Mintlify7:47
  9. MTS · · 6 min

    Good docs need structure, even as agents take over writing

    “agents are the primary writers and maintainers of content now.”
    Han WangCo-founder, Mintlify16:08
  10. Latent Space · · 7 min

    Measured chemical landscapes let agents run thousands of campaigns

    “last weekend, we pulled out over 10,000 of them and we're not even trying that hard.”
    Ian QuigleyCo-founder & CEO, Leash Bio1:21:47
  11. The Pragmatic Engineer · · 6 min

    Production replay lets agents validate fixes against real traffic

    “essentially you can capture real production traffic with just one middleware because now we have standardized language frameworks across all the services”
    Uday Kiran MedisettyDistinguished Engineer, Uber1:07:36
  12. The Pragmatic Engineer · · 5 min

    Minion handles about 15% of Uber’s production PRs

    “around 15 percent of all of the production PRs go through go through minion right now”
    Uday Kiran MedisettyDistinguished Engineer, Uber50:51
  13. The Pragmatic Engineer · · 6 min

    Uber still requires human review on every PR

    “we want to err always on the safer side because the cost of an incident for Uber is like really high.”
    Uday Kiran MedisettyDistinguished Engineer, Uber59:26
  14. TBPN · · 5 min

    One researcher plus AI agents found a Chrome full-chain exploit in 6 days

    “What that means is that you just point your web browser at one malicious website, and the attacker will own your entire Windows PC.”
    Oege de MoorFounder & CEO, XBOW2:00:37
  15. TBPN · · 5 min

    How pooled compute lets startups and public users share GPU capacity

    “the benefit of pooling is we get to essentially load balance across these different customer types”
    Anjney MidhaFounder, AMP55:49
  16. Decoder · · 5 min

    Enterprise users tolerate brittle agents; consumers walk away

    “What I know about consumers is that the first time it doesn't work, they just walk away. They have no incentive to try again to overcome frustration or brokenness.”
    Nilay PatelEditor-in-chief, The Verge19:49
  17. AI & I · · 4 min

    Why every personal agent at Every died and one company agent won

    “there's all the people in the org who don't really. We don't really want to maintain their own cause and then the people who do want to do it like eventually you kind of get tired of it”
    Dan ShipperCo-founder & CEO, Every13:38
  18. The Cognitive Revolution · · 7 min

    Model approval can stall enterprise AI agents

    “it's sitting in a compliance backlog today somewhere in their organization where it's pending the approval”
    Woodson MartinCEO, OutSystems14:03
  19. The Pragmatic Engineer · · 5 min

    Run an AI software factory on one module, not the whole system

    “this area here is where we're trying a full software factory approach, where we don't look at the code anymore, but we specify the interface.”
    Sam NewmanIndependent consultant & author1:47:06
  20. Sabrina Halper Show · · 5 min

    Wiz Atlas makes vulnerability research repeatable

    “Maybe now, in a way, I don't need to be a vulnerability researcher to do vulnerability research.”
    Yinon CosticaCo-founder, Wiz20:20
  21. a16z Podcast · · 8 min

    Cyber agents need domain expertise and layered safety checks

    “we tested the open weight ones and the most advanced closed models, they all found eight.”
    Kevin MandiaFounder & CEO, Armadin30:17
  22. Grit · · 3 min

    Vercel runs support, BDR and RevOps with agents

    “the remaining support team that we have really is now finding legitimate problems in the product and then working really closely with the engineering”
    Jeanne GrosserCOO, Vercel58:16
  23. Lenny's Podcast · · 2 min

    A Dot warned Tibo of a production outage before a live demo

    “The fundamental difference in the way that we've built dots is the harness does not run on the machine.”
    Tibo SottiauxHead of ChatGPT, OpenAI22:02
  24. Behind the Craft · · 3 min

    Build trust in a bot by turning observed runs into skills

    “I think one thing I always try to encourage people to do, you know, whether you're a pstack user or not is to actually spend the time to sort of observe the bot or agent doing”
    Lauren TanLead Engineer, SpaceXAI39:50
  25. Behind the Craft · · 5 min

    An eng lead bot that only delegates lets one person run agent swarms

    Lauren has an eng lead bot that never does work itself and only delegates to engineer bots, which spawn cloud agents. Her agent-friendly codebase means PRs sometimes auto-merge before she reads them.

Not yet2

  1. The MAD Podcast · · 7 min

    RLHF made models pleasing, but raised automation expectations

    “The root of the problem to me is that RLHF actually greatly increased expectations.”
    Diogo AlmeidaCEO, TypeSafe33:27
  2. The MAD Podcast · · 6 min

    Copying Jev’s interface does not reproduce its reliability

    “the problem is it was really really clearly not robust and that was something that has like plagued me for years”
    Diogo AlmeidaCEO, TypeSafe54:03

Each moment is a few minutes of a long interview, cut by Real Alpha and filed under how it answers by a model that read its claim and quote. Open a moment for its full transcript. Last moment added Oct 23, 2026.