Tadyo

I test AI tools so you don't have to
3 real picks/week → The Lever newsletter
No hype. Just tools that move the needle
🔗

[Tadyo @ucfmkr]

Hi, I’m testing https://t.co/BiLnVehz9E for finance teams that flags supplier price increases, duplicate invoices & invoice discrepancies.

I’m looking for a few finance managers to show it to for 5mn and get feedback. No payment required.

Would you be open to taking a look?

[Tadyo @ucfmkr]

@polydao this is the line that compounds

every team I see plateauing on AI tooling outsourced understanding too early and built brittle workflows

[Tadyo @ucfmkr]

@_vmlops the terminal-native angle matters more than people give credit for

every context switch back to a chat tab is a place you lose the thread

[Tadyo @ucfmkr]

@zostaff 20 minute build, 18 month compliance review

the part of a quant's job that's hard to replace is sitting in front of the regulator explaining why the model did what it did

[Tadyo @ucfmkr]

@DamiDefi in fintech we ended up bolting a separate evaluator agent on top whose only job is to flag risk patterns the main agent might miss

not elegant. but it lets us actually deploy

[Tadyo @ucfmkr]

@rileybrown @Rasmic for teams already running agents in production this year is just compounding

the gap between teams shipping agents and teams talking about them is the actual story of 2026

[Tadyo @ucfmkr]

@arrakis_ai time-based stops work better than task-based for us

the agent will keep filing tickets to itself if you let it judge its own done state

[Tadyo @ucfmkr]

@kapilansh_twt we hit this in our enterprise eval

'extended' is a marketing word. cursor's the only one that made me ask sales for the actual number

[Tadyo @ucfmkr]

@rileybrown the part that took us 3 months to figure out was running multiple codex sessions in parallel against the same repo

once we let go of the 'one developer one session' habit our throughput doubled

[Tadyo @ucfmkr]

@KanikaBK 37x faster on what task. my experience is these speedups disappear the moment your codebase has any real legacy weight

the boring measurement is what you ship per week, not what your assistant feels like

[Tadyo @ucfmkr]

@svpino esc+esc to checkpoint changed how my team uses agents. we let it try riskier things now because the undo is real

the cultural shift was bigger than the productivity one

[Tadyo @ucfmkr]

@RoundtableSpace speed is rarely the bottleneck on production scraping

brittleness is. the same site shipping a class name change is what wakes you up at 3am, not the runtime

[Tadyo @ucfmkr]

@GoogleAIStudio an internal compliance dashboard that pulls from 3 systems we have not connected before

the boring kind nobody tweets about

[Tadyo @ucfmkr]

@cyrilXBT this is real but the framing is off. distribution was always the bottleneck

AI just made it more obvious because the build excuse went away

[Tadyo @ucfmkr]

@RoundtableSpace local is a strong story for compliance and privacy. nowhere close on iteration speed

most teams pick the hosted tool because they ship faster, not because they trust the cloud more

[Tadyo @ucfmkr]

@AlexFinn ran /goal overnight on a payments reconciliation script

the number of edge cases it surfaced that we had not thought of was uncomfortable

[Tadyo @ucfmkr]

@thsottiaux the instruction quality bottleneck is real

we found 80% of our /goal failures were us being lazy about the brief, not the model being limited

[Tadyo @ucfmkr]

@meta_alchemist the archive vs delete distinction is the part most teams miss

we lost 2 days last month restoring a context file someone yeeted thinking it was stale

[Tadyo @ucfmkr]

@BoringBiz_ the layer with the worst unit economics right now is the application layer

inference costs are dropping and so are switching costs. nobody has a moat there yet

[Tadyo @ucfmkr]

@Rixhabh__ the part that landed for our team was treating context as a first-class artifact

most of our prompt failures were actually context failures we hadn't named