Tasklet agents are now 30% more cost-efficient

We’ve been working hard to make your Tasklet agents smarter and more credit-efficient — so you can do more with the same spend. We are committed to giving you the most value for every dollar you spend on AI, and over the past couple months we’ve rolled out a number of changes to make that happen.

Here’s a look at the tech behind a few of our recent improvements:

Deferred Tool Calling

Your agent loads the tools it needs when it needs them, instead of carrying its whole toolbox into every conversation.

We reduced the always-on set of tools down to a small core and let agents discover additional tools only when needed. The number of tools an agent can use no longer makes every request more expensive. For organizations with a growing number of agents and connections, this makes a massive difference. Each agent your team uses is now capable of very complex workflows without being any more expensive to use for everyday tasks.

This went into production in June and is averaging about 15% savings to our customers.

Smart Delegation Across Models

Your Tasklet agents can now break complex work into focused tasks and hand them off to dedicated subagents — picking the right model and level of intelligence for each one.

Agents can now delegate to faster, lower-cost models for routine work, even to models from other labs. With Tasklet, your team reaps the benefits of all the best models without having to switch between different AI apps. You don’t have to tell your agent that the cost of GPT-5.6 Luna just went down by 80% — it already knows, even if it’s running on Opus 5. Your agent decides when to delegate, and you’ll see it as it works. Some steps get finished by a cheaper, faster model while your main agent stays in control at the level you picked.

Code-Based Tool Calling

Code-based tool calling lets your agents make a plan once and then use code to execute the repetitive work at scale — which is faster, more reliable, and more credit efficient.

Instead of an AI model repeatedly reasoning through the same steps, your agent can write a small program to loop through records, filter results, rename files, process data, or coordinate many tool calls. The agent provides the judgment and direction, then uses code to execute the repetitive work.

This has only been live in Tasklet for about two weeks, but the early feedback is already meaningful: one customer told us that two of their main agents are now running about 40% cheaper than before.

In Summary

All of these improvements make your agents faster, more reliable, and more affordable to run — and this is just a sample of our efforts to drive down costs for you. We are relentless in working to make Tasklet the place where you get the most value for every dollar you spend on AI.

AI budgets are easy to spend — we want this one to be easy to justify. While you’re working on your business, we’ll be here working to improve the agents your business runs on.

Have questions? Drop me a line at hiranya@tasklet.ai