PhilLimmer

You don't need AGI to flag a duplicate invoice but most orgs are paying frontier prices for it anyway. Akai allows you to route the right workflow to the right model for the task, and has backup options for when one provider is having an outage. Seems obvious but not many people are not doing this effectively and are burning tokens in the process - Akai makes it easy.

@OurielOhayon @Bouazizalex @get_akai @eToro just messaged you on LinkedIn

@Bouazizalex @get_akai @pjl_92 @eToro let s

1,040,000 cases automated by @get_akai since we launched. It's exponential, not linear! Worth pointing out - a case isn't a chat session. It's an agent starting a workflow, working through the steps, handling edge cases, and finishing the task end to end - no human picking up the slack halfway through. We’re just getting started!

@OurielOhayon @Bouazizalex @get_akai @OurielOhayon we've done a number of agents for crypto partners too - wallet risk scans, asset transfer requests, transaction monitoring across tools like Fireblocks, Hypernative, Ellliptic etc.

@OurielOhayon a lot of finance to start with - payment processing running as a background task, settlement reconciliation, NetSuite expense syncing, invoice adjustments. but it spans so much further now: legal (contract redlining), HR (onboarding, right-to-work checks, gov portal filings), IT (asset/order tracking), security (pen-testing etc) and support (tier-1 tickets end to end). All of these have complex logic and branching in place with long-running agent loops. biggest inflection was when people started embedding agents directly into external UIs via our Akai APIs - that's when growth went exponential. would be happy to show you a demo

@shaharkaminsky a lot of finance to start with - payment processing running as a background task, settlement reconciliation, NetSuite expense syncing, invoice adjustments. but it spans so much further now: legal (contract redlining), HR (onboarding, right-to-work checks, gov portal filings), IT (asset/order tracking), security (pen-testing etc) and support (tier-1 tickets end to end). All of these have complex logic and branching in place with long-running agent loops. biggest inflection was when people started embedding agents directly into external UIs via our Akai APIs - that's when growth went exponential.

@Bouazizalex @get_akai What are some examples? Both mundane and complex would be interesting to hear!

@OurielOhayon @Bouazizalex @get_akai Many finance operations - transaction matching, reconciliations, KYC, payouts, expenses etc. across Netsuite, SAP, banks, internal systems and any other systems people use - and across legal, support, marketing, talent, hr we’ve seen huge demand

A large chunk are finance operations - transaction matching, reconciliations, KYC, payouts, expenses etc. across Netsuite, SAP, banks, internal systems and any systems people use Many finance workflows tend to be crazy manual but also have a lot of complex logic - akai handles it But honestly across legal, support, marketing, talent, hr we’ve seen huge demand

we follow this general structure at @get_akai and implement it for our clients to have the granular control they need where we've seen success: - model routing built into the agent layer, not bolted on after - caching orientation knowledge across runs so agents don't rediscover the same context twice - agents that self-optimise - flagging redundant tool calls, stripping what isn't needed, getting cheaper every run we're seeing 70-80-90% cost reductions on our long-running complex agents

How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching. Better Defaults (not Usage Caps) – Engineers can choose any model they want, but defaults matter. We’re experimenting with defaulting to open weight models like GLM 5.2 and Kimi 2.7 through our LLM gateway, while still encouraging engineers to choose the right model for the task. 91% of our employees were never hitting their usage caps, so instead of lowering caps and driving up alerts, we're moving to cheaper defaults. Note that code reviews use a diversity of models, so they can check each other's work. Better Routing – In our custom harnesses, we preprocess prompts and route to the best model for the job, considering cache hits and model pricing. For instance, you may want a frontier model for planning, but not for execution where they can be overkill. Ultimately, humans shouldn't be choosing models - AI can automate this task. Better Caching – Cache misses are the easiest way to drive your cost up. All of our requests are cache aware, so we’re reusing a warm cache wherever possible. For example, our cache hit rate went from 5% → 60% in LibreChat once properly implemented. Keep Context Lean – Start fresh sessions when switching tasks. Scope file context narrowly. Disconnect unused tools. Don't just compact. The goal isn't fewer tokens used, it's fewer tokens wasted. Better Visibility – Our engineers can use as many tokens as they want, from whatever model they want, but we’ve made usage visible – and the more you spend on AI, the more impact we expect. The goal isn't to suppress usage. It's to build the infrastructure that makes exponential growth sustainable. Putting this into practice has cut our AI spend nearly in half, while our token usage continues to grow.

@0xMorlex Not sure he can argue too much on the trust front rn

ANTHROPIC’S CEO RUNS A NEARLY $1T AI COMPANY. HIS PREDICTION: 50% of entry-level white-collar jobs could disappear within 1-5 years. then he explains why he left OpenAI: “why argue with someone when you don’t trust them?” Dario Amodei is building one of the most powerful technologies in history while openly warning that it could erase millions of jobs. his message: - AI will automate the execution - human judgment becomes the final moat - governments are moving slower than the models - waiting for disaster before regulating is too late the person racing to build the future is also warning us what it may destroy.

@ridd_design @meaghaneschoi Not convinced from what I’ve seen so far - initial design is great from 0->1, but it fails on iterations and everything merges into the same fonts/layers/animation styles Also wtf does “do design” even mean 😭 TASTE IS EVERYTHING!

"I think our models will be good enough by the end of the year to do design" @meaghaneschoi from Anthropic shares what that means for designers 👇

@Tristanrhee3 Pet robots

What's coming after artificial intelligence?

@MatthewBerman Really? Just like their agent workflow builder was gonna kill n8n… that aged well

OpenAI might swallow up every piece of software eventually.

@perplexity_ai hmm... isn't this also going to ramp up token usage and bloating the context on each run?

Introducing Brain in Computer. Brain is a continuously learning memory system. Every task on Computer plugs into a context graph built by Brain. It makes Computer more stateful with every run. Available as a research preview for all Perplexity Max subscribers.

although working with AI kinda feels like being on a really fast rollercoaster that I can't decide if i'm enjoying or it's making me feel sick

the last 6 months i've spent travelling the world and building Akai both fun, both exhausting, both I intend to continue kinda like the combination of ai agents and tropical beaches

87% cost reduction on complex, long-running agents... that's the impact so far of Akai's new feature (called Optimise) run 1: $ 1 run 30: $ 0.13 more details in the 🧵

9 months ago, @deel’s @pjl_92 sent a message to a small group of some of my smartest colleagues: “Get on a plane. Be in London tomorrow.” I was lucky enough to be invited. For the next 3 days, we sat in a WeWork from 9am to 9pm with one of the world’s leading AI experts to learn from. We built agents and pushed the boundaries of the workflows. We asked every question imaginable. What can agents do? What can’t they do? How far can we push the limits with current capabilities? At the time, it felt like an experiment. Today, it doesn’t. Fast forward 9 months: Deel launched @get_akai, our agentic processing platform. Every operations team now uses it. Processes that once required entire teams are increasingly being handled by software. Not because people became less valuable. Because the economics changed. The old formula: Headcount cost × number of people required = problem solved The new formula: Token cost + oversight + initial build = problem solved If you work in operations, consulting, outsourcing, support, finance, HR, payroll, or professional services: Pay attention. This isn’t a future trend. It’s happening now. Don’t wait for permission. Start experimenting. The companies taking similar actions today will look unrecognizable 6 months from now. And the gap between them and everyone else is growing every week.

right you need to move to SF but you don't need to waste money on fancy restaurants etc. so for ai the answer isn't using fewer tokens, it's using the right tokens and the right models. cache orientation knowledge, not raw context, route simple tasks to cheaper models, strip AI from sections that don't need it, know which agents are actually performing so you need visibility into every run, every tool call, every decision the agent made

YC Managing Partner @harjtaggar says many young founders are still too conservative with token spend and advises them to not skimp on token budget. "Token spend is more like rent. You don't say, 'I'm not moving to San Francisco to build my tech startup because the rent's expensive.' Yes, it's expensive, but it's the place to be. You don't conserve on that. You actually want to spend money on that. There's other things you save money on." "Many of the founders, especially if they're younger, just don't have the budgets to use the latest models and be cost insensitive and token-max. They're just used to conserving and trying to do more with less." "Internally, we wonder if there's some training to push the founders to spend more on tokens because maybe they're thinking about tokens as something you should conserve spend on."

I think the teams that win on AI cost now aren't the ones who token max or the ones who restrict usage, but the ones who know exactly what every token is doing. (we built this 👉 @get_akai)

this requires infrastructure that gives you granular control - not just a spend cap, but visibility into every run, every tool call, every decision the agent made.

I agree, but it's easy to say if you have the money to spend. the problem is most teams don't know where their tokens actually go. they're now seeing the bill and panicking - but it's important to understand if it's driven by redundant tool calls, bloated context, or sections where AI isn't doing anything useful.