MEGA Code (@megacode_ai) header
MEGA Code (@megacode_ai) profile picture

Enabling self-evolving agent systems through optimization. GitHub: https://nitter.cf/t.co/ywCwO95zKF

Joined February 2026
Photo by MEGA Code (@megacode_ai)Photo by MEGA Code (@megacode_ai)Photo by MEGA Code (@megacode_ai)Photo by MEGA Code (@megacode_ai)Photo by MEGA Code (@megacode_ai)Photo by MEGA Code (@megacode_ai)Photo by MEGA Code (@megacode_ai)Photo by MEGA Code (@megacode_ai)Photo by MEGA Code (@megacode_ai)Photo by MEGA Code (@megacode_ai)
MEGA Optimus is here. Point it to a project folder. Optimus writes the spec, builds the eval harness, and runs the optimization loop end to end. Reduce your token spend and latency while improving accuracy. megacode.ai/start-optimizati… Autonomous AgentOpt
3
2
12
1,380
South Korea’s public AI project requires participating companies to use models from other Korean developers as well as meeting a domestic model requirement. There’s a good design principle in keeping the model separate from the citizen records, agency rules, and workflows.
3
85
AI could make it practical for governments to build services for communities that were too small to justify custom software. A rural municipality might want a tool that helps residents apply for permits for home renovations, as each district may have its own codes. If local teams can adapt a shared model cheaply enough, each district can get a service that fits without having to fund a model of its own.
3
27
Faster models like Gemini 4 Hydrogen could make agents useful on live support calls, where the answer has to come back while the customer is still on the line. A customer asks why their bill went up. While the rep is talking to them, an agent could compare the last two invoices, check for a plan change, and point to the charge that explains the difference. If it can do that reliably during the call, the rep can answer then and there instead of opening a ticket and getting back to the customer tomorrow.
Introducing Gemini 4 Hydrogen, an extremely capable and fast model built to bring Gemini 4 to everyone! Available to all tiers of Google AI Studio, Gemini API, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro + Ultra subscribers in Google Antigravity.
4
99
A model can be good at picking from a list and still make the wrong call because the right option was missing. A support agent deciding between refunding and denying a claim might need a third option, such as first checking whether the order was delivered. When these decisions are separate calls, it’s easier to see whether the model picked badly or the workflow gave it a bad set of choices. Those need different fixes. That’s where MEGA comes in. It can use the trace to see whether the model picked badly or the workflow gave it the wrong options, then test a fix on both the cases that failed and the ones that already worked.
Today we’re launching the Gemini Decisions API, powered by our new Gemini Decision 1 model in @GoogleAIStudio. Instead of generating text token by token, it’s built for closed-set choices scoring up to 1,000 questions across text, images, audio, video + PDFs in a single pass. A few details: - 3 decision types: yes/no, multiple choice (up to 255 options) + rubric scoring - $0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite) - leads on several benchmark evals with lowest calibration error Google teams are already building with it, below is a look at Stitch Rapid designing a full product flow in seconds (1x speed)
4
128
Autonomy scales, reliability doesn’t. Agents can call tools and take action on your behalf, but catching failures, verifying fixes, and improving them meaningfully is still largely manual. That’s the problem MEGA Code exists to solve.
1
5
68
MEGA Optimus: optimizes the entire pipeline itself. +10.1% aggregate task score over GEPA. Wisdom graph then turns every run into reusable skills. Task success 31.5% to 46.5% on Skillsbench (84 tasks, 11 domains).
1
3
20
If customers use AI to make calls and businesses use AI to answer them, both agents could report that they handled a call while nothing got done. One says it’ll call back. The other says it’s waiting for a callback. The appointment never moves. If both agents log that as a success, there’s no reason for either to change. Mark it unresolved and it becomes a case to test the next version on. Does it confirm a new time, or at least leave someone responsible for following up?
4
44
A country could get more out of sovereign AI by funding local teams to adapt models than by funding a flagship model of its own. The government can help create that industry by becoming an early customer. Then it can give local teams real work in ports, utilities, and public services, and pay for improvements those operators can actually use. All without the dependency on foreign vendors or data leakage risk.
4
37
A company can get value from building a replacement for a piece of software without ever switching to it. AI could shift the leverage companies have during negotiations at renewals.
3
24
Super important for enterprises to distinguish between expensive workflows that failed and experiments that helped employees find better ways to work. To clarify, a spending cap can limit employees from trying things whose value they don’t know yet.
I appreciate the live experiment happening between the world's #1 and #2 companies by market cap: NVIDIA: virtually no limit on AI spend (at $10K/month you need to justify and then it auto approves). AI use encouraged Apple: AI use not really encouraged We'll see what it will mean for business $$$ !
3
75
If a vendor says its product uses a particular AI model, should the customers be able to verify that? A receipt proving which model produced the response would bring much more transparency and trust. It wouldn’t necessarily mean the response is good, but it would establish which model needs to be investigated when something goes wrong.
1
3
85
Keeping Git matters here. An enterprise adopting coding agents already has build systems, permissions, scripts, and habits built around its repositories. Replacing all of that would add another migration before the agents could be useful. Some of the infrastructure needed for AI adoption will involve substantial engineering so everyone else can keep using the tools they already know.
Pando is a Git backend which can scale to very large monorepos. We took the hard road of continuing to use Git, which means we could enable Pando with no change to devex. Considering providing this to others, let me know ([email protected]) if you'd be interested.
4
135
Employees can have different AI interfaces. They still need shared answers about how the company works. If each assistant builds its own memory, one might remember an exception as standard practice. Another might keep using a process that changed last month. Personal memory should help an assistant understand the employee. A shared knowledge base should establish what’s current for the company, with a clear way to resolve conflicts between the two. Our Wisdom Graphs help enterprises navigate these challenges.
4
41
Whose assistant is an AI in a group chat? If one person asks it to book dinner, another mentions a budget, and a third says nothing, it has to decide what counts as agreement. A personal assistant can learn one person’s preferences, but a shared assistant needs a way to handle disagreement without treating the most vocal person as the whole group.
4
92
AI will make it harder for companies to leave certain assumptions untested. A junior dev can build a prototype or an MVP that customers can try, in the matter of days. But will companies welcome that when it comes from someone junior and contradicts a decision they’ve already made?
3
34
A person can approve an AI’s recommendation without noticing how much the AI shaped their opinion. That makes “a human made the final decision” less reassuring on its own. For decisions where independent judgment matters, it may be worth asking people to write down their view before showing them the AI’s recommendation. Otherwise agreement can look like a second opinion when it started with the first one.
3
32